Patent Grant
12114986
Embodiments described herein relates to the field of medical devices, systems and methods and, more particularly, to biometric data capture for event prediction in respect of a monitored environment.
The procedure of identifying threats, adverse events, or any other intraoperative segments of interests in the operating room is a technically challenging task with limitations that both restrict its effectiveness and potential for mitigating risk. Identifying such instances is currently limited to manual video analysis which cannot be performed in real-time, and sensors (e.g., microphones, motion detectors, etc.) in the operating room to automatically recognize key events. The ability of the latter to capture segments of interests is limited, and the former is 1) error prone and not entirely comprehensive, 2) cannot be implemented in real-time (only retroactively), and 3) not scalable. These limitations significantly hinder the potential safety management systems from identifying threats, adverse events, and other intraoperative segments of interest in the pursuit of aiding surgical teams both in real time and post-operatively to improve performance, lower prevalence and mitigate severity of adverse events, and improve patients safety.
Capturing biometric data from individuals (such as healthcare providers) in the operating room using wearable technology and synchronization with various structured and unstructured data streams will assist with automatic detection of events of interest in the operating room in both real-time and post-operatively. The captured biometric data can be utilized to automatically modify the operation of a hospital environment (e.g., a “black box” capturing event information relating to a medical procedure taking place within an operating theatre) that has been augmented (e.g., supplement, augment) with sensors.
Experimental validation was conducted in respect of using sensors to obtain data sets that can be utilized as proxies for measuring stress levels or other physiological conditions in practitioners, and these data sets can be utilized, in some embodiments, to address technical problems associated with finite resources by automatically shifting resource allocation based on the information captured in the data sets.
While experimentation was conducted using a shirt having embedded sensors tracking electrocardiograph data and converting the data into heart rate variability data sets, other variations and types of biometric data sets are contemplated. The sensors are adapted to record, in an embodiment, audio and/or visual data which can be time-synchronized with the captured biometric data such that a recording data object can be established or encapsulated for downstream processing and analysis. The captured biometric data can be processed to generate derivative analytic values (e.g., electrocardiography data to heart rate variability data values), which are then time-synchronized to timestamps of captured video or audio data streams to identify, abnormality-related durations of time during which the data values are greater or lower than a pre-defined threshold data value, or deviate from expected and/or observed distributions.
A technical challenge that arises with equipping the environment with sensors is the volume of data that is generated. A significant volume of data is captured and there are limited computer processing resources and networking resources available. A further limitation is the amount of computer storage space available, which is constrained due to potential healthcare data policies requiring enhanced security, redundancy, uptime, and disaster recovery capabilities.
As described herein in various embodiments, biometric data is utilized to enhance (e.g., control, modify) the operation of the recording mechanism coupled to the hospital environment. The biometric data, such as electrocardiography data, electromyography data, electroencephalography data, electrooculography data, foot/shoe sole pressure data can be processed in real or near-real time to identify data artifacts indicative of potential abnormal (e.g., elevated, lowered) stress levels or abnormalities in the biometric data. These data artifacts, for example, can then be tracked to durations of time which may be regions of interest (or regions for discarding) in respect of future analysis or data processing. These durations of time can be utilized for improved targeting of limited resources, so that the recordings can indicate or automatically be technically improved (e.g., higher resolution, more recordings, reduced compression loss) for a limited duration of time so that improved accuracy and/or fidelity in recordings can be established without requiring significant infrastructure improvements to be made.
In particular, the biometric data can be appended to audio and video stream data, for example, in the form of metadata or a separate time-synchronized stream of information. In some embodiments, the raw biometric data may be transformed into a derivative format, such as heart rate variability values prior to appending to the audio and video stream data. For downstream processing and analysis of the audio and video stream data, the appended biometric data or derivatives thereof may be utilized as an input, for example, into a machine learning model for incident prediction, outlier detection, or surgical skill assessments, among others.
In another embodiment, the appended biometric data or derivatives thereof are processed by a proximate, on-premises system, or a remote cloud-based server, in real or near-real time to generate region of interests corresponding to durations of time associated with the time-synchronized biometric data and corresponding audio and video data streams. The proximate, on-premises system, for example, may include a physical server that is coupled to or physically located within the local sensorized healthcare environment that is adapted for an initial processing of the time-synchronized biometric data and corresponding audio and video data streams.
The proximate, on-premises system may, in some embodiments, be configured to control one or more characteristics of the operation or the output from the local sensorized healthcare environment to a backend data center or centralized data processing server.
In a first embodiment, the proximate, on-premises system is configured to increase (or decrease) a level of recording intensity automatically upon triggering of the beginning (or end) of a duration of time corresponding to a region of interest (e.g., through the detection of an abnormality in biometric data). In this situation, the proximate, on-premises system may control one or more recording devices to increase (or decrease) performance through modification of various recording parameters. For example, an enhanced resolution, reduced compression in recordings, or activation of other data streams (e.g. more cameras, microphones, sensors) during the duration of time may be established such that more information is captured at the expense of downstream processing complexity and storage requirements. In this duration of time, enhanced bandwidth resources may also be provisioned or prioritized to handle the larger flow of recorded data.
In another embodiment, the proximate, on-premises system is configured to request enhanced or prioritized downstream processing resources allocated to the processing of any data that is during or temporally proximate to the duration of time in which the abnormality was detected.
Downstream processing resources may generate a data file storing annotations useful for downstream automated analysis, or downstream human analysis, such as timed bookmarks, rendered subtitles, among others. During downstream automated analysis, a machine learning model may utilize the annotations (e.g., the flagged data) as an input feature such that additional emphasis may be placed on annotated sections during the generation of output logits or normalized predictions.
In a variant embodiment, the modified capture of the video or audio data streams includes modifying an encoding approach (e.g., a compression approach) applied to the video or audio data streams prior to transmission to the centralized computer server. The raw video or audio data streams during the abnormality-related durations of time can be compressed differently such that a higher fidelity to the original data is maintained. This can be implemented through a modification of encoding characteristics, such as a level of lossy compression, the selective use of loss-less compression, the change in compression codec, a change in compression type (e.g., switching between or selectively applying discrete cosine transforms and motion compensation), increasing spatial or temporal redundancy, among others.
In this variant embodiment, the objective is to provide the centralized server with a greater set of features for analysis at the cost of increased downstream processing requirements and/or networking resources for transmission. To limit this potential increased demand on the system, limiting the use of or the requirement for the increased resource can be limited only to the durations of time where abnormalities are detected in the biometric data, and some durations of time temporally proximate to the abnormalities (e.g., +/−1 second, 2 seconds, 3 seconds, 4 seconds, 5 seconds).
In an aspect, there is provided a system for providing alerts of pending events in medical procedures. The system comprises a processor, and a memory storing instructions which when executed by the processor configure the processor to capture biometric data of a healthcare provider during a medical procedure, determine an elevated stress level based on the biometric data, and send an alert based on the elevated stress.
In another aspect, there is provided a method of providing alerts of pending events in medical procedures. The method comprises capturing biometric data of a healthcare provider during a medical procedure, determining an elevated stress level based on the biometric data, and sending an alert based on the elevated stress.
In some embodiments, the biometric data comprises heart rate variability measure of the healthcare providers during the medical procedure.
In some embodiments, the biometric data comprises at least one of electrocardiography (ECG) data to assess the state of the autonomic nervous system and general cardiovascular data, electromyography (EMG) data to capture muscle fatigue, electroencephalography (EEG) to capture brain function, electrooculography (EOG) data and other methods to capture eye tracking, and/or “intelligent” shoe soles to assess the level of fatigue.
In some embodiments, the alert is at least one of a tactile alert, a visual alert, or an audible alert.
In some embodiments, the biometric data comprises ECG readings of a healthcare provider, and the method comprises converting the ECG readings into heart rate variability (HRV) values, determining that an HRV value passes a threshold value, and sending the alert based upon the passing of the HRV threshold value.
In another aspect, there is provided a non-transitory computer readable medium storing instructions which, when executed by a processor, configure the processor to capture biometric data of a healthcare provider during a medical procedure, determine an elevated stress level based on the biometric data, and send an alert based on the elevated stress.
In another aspect, there is provided a non-transitory computer readable medium storing instructions which, when executed by a processor, cause the processor to capture biometric data of a healthcare provider during a medical procedure, determine an elevated stress level based on the biometric data, and send an alert based on the elevated stress.
In another aspect, there is provided a system for predicting an event in a medical procedure. The system comprises a processor, and a memory storing instructions which when executed by the processor configure the processor to capture biometric data of a healthcare provider during a medical procedure, determine an abnormality in the biometric data, and generate models for use in the study of operating procedures based on the abnormality.
In another aspect, there is provided a method of predicting an event in a medical procedure. The method comprises capturing biometric data of a healthcare provider during a medical procedure, determining an abnormality in the biometric data, and generating models for use in the study of operating procedures based on the abnormality.
In another aspect, there is provided a non-transitory computer readable medium storing instructions which, when executed by a processor, cause the processor to capture biometric data of a healthcare provider during a medical procedure, determine an abnormality in the biometric data, and generate models for use in the study of operating procedures based on the abnormality.
In another aspect, there is provided a system for collecting and processing medical or surgical data. The system has a plurality of hardware units for collecting real-time medical or surgical data streams having a control interface coupled by a network to cameras, sensors, audio devices, and patient monitoring hardware, the real-time medical or surgical data streams relating to a real-time medical procedure within an operating or clinical site. The data streams include biometric data from healthcare providers during the medical procedure. The system has an encoder with a network server for synchronizing and recording the real-time medical or surgical data streams to a common clock or timeline to generate a session container file. The network server configured to control a multi-nodal perception engine to: generate a protocol for data extraction from the session container file; process the data using the protocol to extract patterns for time-stamped clinical events within the session container file, each time-stamped clinical event associated with a confidence level; generate an interface indicator for a temporal sequence of the time-stamped clinical events within the session container file and error assessments, the interface indicator identifying each of the time-stamped clinical events and the associated confidence levels; generate a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures with neural networks for modelling correlation of data for interference and feature extraction.
In some embodiments, the device middleware and hardware to establishes a secure reliable connection using a network infrastructure for communication with the encoder and the hardware units, the device middleware and hardware for translating, connecting, and formatting the real-time medical or surgical data streams received independently from the hardware units.
In some embodiments, the device middleware and hardware implements data conformity and accurate synchronization for the real-time medical or surgical data streams using network protocols for clock synchronization between the hardware units to assist the encoder to generate the session container file.
In some embodiments, the encoder and device middleware and hardware are operable to interface with third party devices to receive additional data feeds as part of the real-time medical or surgical data streams.
In some embodiments, a central control station accessible using the control interface, the control station configured to control processing of the data streams in response to input control comprising play/pause, stop session, record session, move to session frame, split-display, recording status indicator, and log file.
In some embodiments, network infrastructure provides increased fail-over and redundancy for the real-time medical or surgical data streams from the hardware units.
In some embodiments, a storage area network for storing data container files of the real-time medical or surgical data streams until scheduled transmission.
In some embodiments, the encoder implements identity anonymization and encryption to the medical or surgical data.
In some embodiments, the encoder processes the real-time medical or surgical data streams to generate measurement metrics relating to the medical procedure.
In some embodiments, the real-time medical or surgical data streams correlates to a timeline, wherein the encoder detects events within the real-time medical or surgical data streams at corresponding times on the timeline, and tags and timestamps the session container file with the events, the timestamps corresponding to times on the timeline.
In some embodiments, an intelligent dashboard interface for annotation and tagging of the synchronized medical or surgical data streams, wherein the intelligent dashboard may implement a viewer with playback viewing for reviewing content and interface controls for tagging content.
In some embodiments, the intelligent dashboard is multi-dimensional in that the union of all dimension variables for the medical procedure may indicate a specific set of one or more applicable annotation dictionaries or coding templates.
In some embodiments, example variables that may be used to determine the annotation and tagging dictionary may be: the type of medical procedure being performed, the aspect of the procedure that is being analyzed, the geographic area/region where the procedure is being performed.
In another aspect, there is provided a multi-channel encoder for collecting, integrating, synchronizing and recording medical or surgical data streams onto a single interface with a common timeline or clock, the medical or surgical data streams received as independent real-time or live data streams from a plurality of hardware units, the encoder having a network server for scheduling transmission of session file containers for the recordings, the encoder processing the medical or surgical data streams to generate measurement metrics relating to a real-time medical procedure, the encoder configured to generate a protocol for data extraction from the session container file; process the data using the protocol to define patterns for time-stamped clinical events within the session container file; generate an interface indicator for a visual sequence of the time-stamped clinical events within the session container file and correspondence assessments; generate a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures.
In some embodiments, the encoder generates as output a single session transport file using lossless compression operations.
The encoder detects completion of a recording of the data streams and securely encrypts the single transport file.
In some embodiments, the encoder implements identity anonymization to the medical or surgical data.
In some embodiments, the data streams comprising audio, video, text, metadata, quantitative, semi-quantitative, and data feeds.
In another aspect, there is provided a method for collecting and processing medical or surgical data comprising: receiving, at a multi-channel encoder, a plurality of live or real-time independent input feeds from one or more data capture devices located in an operating room or other patient intervention area, the input feeds relating to a live or real-time medical procedure; synchronizing, by the encoder, the plurality of live independent input feeds onto a single interface with a common timeline or clock; recording the synchronized input feeds using a network server; generating, by the encoder, an output session file using the synchronized input feeds; transmitting the output session file using the network server; generating a protocol for data extraction from the session container file; processing the data using the protocol to define patterns for time-stamped clinical events within the session container file; generating an interface indicator for a visual sequence of the time-stamped clinical events within the session container file and correspondence assessments; and generating a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures.
In some embodiments, the method may involve processing the data streams for identity anonymization.
In some embodiments, the method may involve routing the data streams using a switch router to the encoder.
In another aspect, there is provided a cloud based system for collecting and processing medical or surgical data comprising: an encoder having a control interface for, in response to receiving a control command, triggering collection of real-time medical or surgical data streams by smart devices including cameras, sensors, audio devices, and patient monitoring hardware, the medical or surgical data relating to a real-time medical procedure within an operating or clinical site, the encoder for authenticating the smart devices, the smart devices synchronizing the real-time medical or surgical data streams by embedding timestamp markers within the real-time medical or surgical data streams, the timestamp markers generated by each smart device by a device clock; a media management hub server with middleware and hardware for translating, connecting, formatting, and recording the real-time medical or surgical data streams to generate session container files on network accessible storage devices; wireless network infrastructure to provide a secure network connection between the encoder, the smart devices and the media management hub server for communication of the real-time medical or surgical data streams; a central content server for storing and distributing the session container files and providing a two-way communication interface for the media management hub to implement a file transfer handshake for the session container files. The central content server is configured to: generate a protocol for data extraction from the session container file; process the data using the protocol to define patterns for time-stamped clinical events within the session container file; generate an interface indicator for a visual sequence of the time-stamped clinical events within the session container file and correspondence assessments; generate a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures; and switching or gateway hardware to transmit the session container files from the media management hub to the central content server.
In some embodiments, the media management hub server broadcasts clock data to the smart devices for synchronization of the device clocks.
In some embodiments, the encoder provides a user interface to receive the control command and display real-time visual representations of the medical or surgical data.
In some embodiments, the media management hub server aggregates, packages, compresses and encrypts the real-time data streams to generate the session container files.
In some embodiments, the media management hub server manages the smart devices based on location, schedule, zone and requirements.
In some embodiments, the media management hub server receives operating status data from the smart devices to generate a management interface with a visual representation of the operating status data for the smart devices, the operating status data including online, offline, running capture, and on-board storage.
In some embodiments, the media management hub server processes the operating status data to detect smart devices operating outside of normal conditions and in response generating an alert notification of the detected smart devices operating outside of normal conditions.
In some embodiments, the media management hub server implements a device communication interface for the smart devices to implement a device data transfer handshake for the real-time medical or surgical data streams.
In some embodiments, the media management hub server authenticates the smart devices.
In another aspect there is provided a computational intelligence platform for receiving the session container files to construct an analytics model to identify clinical factors within the session container files for predictions, costs and safety hazards, the analytics model providing a network for extracting features, correlations and event behaviour from the session container files that involve multivariable data sets with time-variant parameters.
In some embodiments, a training or education server may receive the session container files, process the session container files to identify root causes of adverse patient outcomes and generate a training interface to communicate training data using the identified root causes and the session container files.
In some embodiments, the smart devices include motion tracking devices for markerless motion tracking of objects within the operating or clinical site, the system further comprising a processor configured to convert captured motion data from the motion tracking devices into data structures identifying human factors, workflow design and chain-of-events.
In some embodiments, the time-stamped clinical events within the session container file is stored with associated metadata for duration and frequency of each time-stamped clinical event.
In some embodiments, the network server or encoder uses patterns for time-stamped clinical events within the session container file to identify and extract features from the session container file for correlation or spectral analysis based on temporal nature of the time-stamped clinical events within the session container file.
In some embodiments, the network server or encoder identifies frequent temporal events as patterns leading to adverse events or errors in the timeline and develops predictive models to identify critical events during the real-time medical procedures.
In some embodiments, the network server or encoder groups the time-stamped clinical events within the session container file into technical and non-technical events.
In some embodiments, the interface indicator comprises an audio representation of the sequence of the time-stamped clinical events. In some embodiments, the interface indicator comprises a video representation of the sequence of the time-stamped clinical events.
In some embodiments, the network server or encoder configures the multi-nodal perception engine for filtering the time-stamped clinical events within the session container file using machine learning with feature extraction for event correlation using computational intelligence, the multi-nodal perception engine interfacing with distributed hardware units.
In some embodiments, the network server or encoder generates the predictive data model by generating an Attribute-Relation File Format and the artificial intelligence network data structures and creates the temporal sequence of the time-stamped clinical events within the session container file, the temporal sequence relating to predicted locations of surgical error, wherein the interface indicator receives rating indicia for the error assessments of the time-stamped clinical events.
In some embodiments, the network server implements post-processing of the time-stamped clinical events within the session container file for comparative processing for outcome links to generate hazard maps for the interface indicator.
These are illustrative example embodiments and other variations may be described herein.
Embodiments will be described, by way of example only, with reference to the attached figures, wherein in the figures:
Embodiments of methods, systems, and apparatus are described through reference to the drawings.
Wearable technology and video software may be used to capture biometric data of members of the operating team for the purposes of automating the identification of threats, adverse events, and other intraoperative segments of interest. This would expand the potential for safety management systems to reduce risk for errors and improve outcomes. Significant advances in wearable technology, particularly the ability to reliably measure physiological metrics unobtrusively without wired sensors, afford new opportunities for biometric tracking and data integration.
Increased safety in healthcare procedures requires accurate, actionable healthcare insights, but a countervailing consideration is the rising cost and complexity of delivering quality healthcare service as the population ages and the complexity of procedures increases. Another challenge is that there can often be a time difference between a causative event and adverse events, and computationally intense machine learning data model architectures and algorithms require as much information as possible to be provided as input features. However, an increase in input features leads to increased computational resource requirements.
A computer implemented system for tracking and/or processing data streams captured during a medical procedure is described in various embodiments herein, augmented by captured biometric data. The system is adapted to provide a technical solution by automatically increasing resource usage intensity during limited durations of time by processing captured biometric data as a proxy indicative of various physiological factors associated with a practitioner.
The captured biometric data is utilized to automatically modify aspects of how data streams are processed or tracked, and in some embodiments, automatically request or provision increased computing or network resources to pre-emptively expand downstream processing capabilities by focusing the limited computing resources around durations of time proximate to abnormal readings in the captured biometric data. In particular, the captured biometric data can be utilized to automatically modify the operation of a hospital environment (e.g., a “black box” capturing event information relating to a medical procedure taking place within an operating theatre) that has been augmented (e.g., supplement, enhanced) with sensors.
A surgeon could, for example, be connected to a biometric device that reads various information associated with the surgeon's physiology. Coupled biometric devices can include heart monitors, oxygen saturation monitors, pressure sensors (e.g., in shoes), brainwave detection scanners, blood pressure monitors, among others. The biometric indicators, for example, may be representative of stress levels. However, as noted herein, stress can both to facilitate and to impair performance in a variety of contexts, and while stress is more likely to impact a task more negatively, it does not always need to do so. Stress may, for example, impact functions such as decision-making, team performance, among others.
Experimentation was conducted in respect of continuous electrocardiogram data for an attending surgeon during surgical procedures to obtain heart rate variability (HRV) measures that were used as a proxy for acute mental stress. Two different measures were used: root mean square of successive differences (RMSSD) and standard deviation of RR intervals (SDNN). Technical surgical performance was assessed on an operating room black box platform using the Generic Error Rating Tool (GERT).
In the context of the operating room black box platform, in some embodiments, the tracked biometric information could be useful as an additional input for a recording system that is used to improve an operating theatre An operating theatre recording system as described in various embodiments herein augmented with biometric data is a useful computer-based tool for improving patient safety and enabling analytical capabilities relating to the recordings (e.g., in real or near-real time, or downstream analysis). The combination of the operating room black box platform and biometric data yields an opportunity to apply data science/machine learning processes to automatically track and train the data science/machine learning processes to model the impact stresses (as tracked using biometric signals as a proxy) for the generation of alerts or other predictive outputs.
The biometric data, whether in a raw or processed format, can be time-synchronized with the recording data. In experimentation, both HRV recording and procedure video recording were time-stamped. Surgical procedures were fragmented to non-overlapping intervals of 1, 2 and 5 min, and subjected to data analysis. An event was defined as any deviation that caused injury to the patient or posed a risk of harm. Such data would allow for identification of segments of interest in real time for the purposes of providing real time warnings and alerts (tactile, visual or audible) when necessary, in addition to offering deeper insights to the state of affairs in the operating room (including physical and psychological states of individuals) for post-operative analysis.
Examples of relevant biometric data include, but are not limited to, electrocardiography (ECG) data to assess the state of the autonomic nervous system and general cardiovascular data, electromyography (EMG) data to capture muscle fatigue, electroencephalography (EEG) to capture brain function, electrooculography (EOG) data and other methods to capture eye tracking, and “intelligent” shoe soles to assess the level of fatigue. Such biometric data may be synchronized with other data streams in the surgical operating room. The sensors record audio and/or visual data which can be time-synchronized with the captured biometric data such that a recording data object can be established or encapsulated for downstream processing and analysis.
The captured biometric data can be processed to generate derivative analytic values (e.g., electrocardiography data to heart rate variability data values), which are then time-synchronized to timestamps of captured video or audio data streams to identify, abnormality-related durations of time during which the data values are greater or lower than a pre-defined threshold data value. The abnormality-related durations of time can be, for example, indicative of periods of stress or depression which can impact adversely on multiple facets critical to optimal performance.
A technical challenge that arises with equipping the environment with sensors is the volume of data that is generated. A large volume of data is captured and there are limited computer processing resources and networking resources available.
A further limitation is the amount of computer storage space available, which is constrained due to potential healthcare data systems policies requiring enhanced security, redundancy, uptime, and disaster recovery capabilities.
The method 2800 comprises capturing 2810 biometric data of a healthcare provider during a medical procedure.
The biometric data can be utilized to enhance (e.g., control, modify) the operation of the recording mechanism coupled to the hospital environment. The biometric data can include data such as electrocardiography data, electromyography data, electroencephalography data, electrooculography data, foot/shoe sole pressure data
At 2820, an elevated stress level can be determined based on the biometric data, and the data can be processed in real or near-real time to identify data artifacts indicative of potential abnormal (e.g., elevated, lowered) stress levels or abnormalities in the biometric data. These data artifacts representing the elevated stress level, for example, can then be tracked to durations of time which may be regions of interest (or regions for discarding) in respect of future analysis or data processing.
As indicated during experimentation, rates of healthcare injury or risk of harm events were significantly higher (47-66 percent higher) in the higher stress quantiles than in the lower stress quantiles for all measured interval lengths using both proxy measures for acute mental stress. The strongest association was observed using 1-min intervals with RMSSD as the HRV measure (P<0.001).
The biometric data can be appended to audio and video stream data, for example, in the form of metadata or a separate time-synchronized stream of information. In some embodiments, the raw biometric data may be transformed into a derivative format, such as heart rate variability values prior to appending to the audio and video stream data.
For downstream processing and analysis of the audio and video stream data, the appended biometric data or derivatives thereof may be utilized as an input, for example, into a machine learning model for incident prediction or surgical skill assessments, among others.
At 2830, the appended biometric data or derivatives thereof can be processed by a proximate, on-premises system in real or near-real time to generate region of interests corresponding to durations of time associated with the time-synchronized biometric data and corresponding audio and video data streams. The on premises system can be a local computer server operating in conjunction with a centralized computer server, which can receive inputs from multiple on premises local systems (e.g., a central data server for a particular facility or ward, or coupled to many facilities having enhanced capabilities for data processing and machine learning data model application).
Based on the detection of an abnormality, the local system of some embodiments is configured to increase (or decrease) a level of recording intensity automatically upon triggering of the beginning (or end) of a duration of time corresponding to a region of interest (e.g., through the detection of an abnormality in biometric data).
In this situation, the proximate, on-premises system may control one or more recording devices to increase (or decrease) performance through modification of various recording parameters. For example, an enhanced resolution, reduced compression in recordings during the duration of time may be established such that more information is captured at the expense of downstream processing complexity and storage requirements. In this duration of time, enhanced bandwidth resources may also be provisioned or prioritized to handle the larger flow of recorded data.
The centralized computer server may be coupled to the local computer server through networked interconnections, and it is important to note that in some embodiments, networking resources can be scarce and need to be conserved, especially as video and/or audio data capture can be very bandwidth intensive. The local computer server for example, may include a physical server that is coupled to or physically located within the local sensorized healthcare environment that is adapted for an initial processing of the time-synchronized biometric data and corresponding audio and video data streams.
The local, on-premises system may, in some embodiments, be configured to control one or more characteristics of the operation or the output from the local sensorized healthcare environment to a backend data center or centralized data processing server. These characteristics can include receiving or generating various alerts generated based on the elevated stress. The alerts can be established in relation to identified durations of time, for example, identified using metadata tags. During these durations of time, an increased level of recording or capture characteristic may be controlled, for example, through increased resolutions, bitrates, modified encoding (e.g., changed compression approaches to reduce loss).
In an embodiment, the centralized computer server is configured for applying one more machine learning models/algorithms to the process received data to generate prediction data objects representative of one or more predicted characteristics or incidents relating to the medical procedure. The prediction data objects, for example, can relate to a level of skill of the practitioner, an expected post-operative outcome (e.g., duration of time to heal), a probability of potential adverse outcome, etc. In some embodiments, the prediction data objects may be generated on a real or near-real time basis to identify specific characteristics that may, for example, lead to serious adverse events, and the centralized computer server may then generate an alert control command data signal that causes an actuation of a tactile alert, a visual alert, or an audible alert. For the durations of time indicative of an abnormal reading, in some embodiments, additional machine vision or other types of analytic algorithms are engaged to conduct additional analyses. In some embodiments, the additional machine vision or other types of analytic algorithms are adapted to conduct analysis on the underlying biometric data to identify features beyond heart rate variability.
Other steps may be added to the method 2800. The biometric data may be captured using wearable technology.
In another embodiment, the proximate, on-premises system is configured to request enhanced or prioritized downstream processing resources allocated to the processing of any data that is during or temporally proximate to the duration of time in which the abnormality was detected.
Downstream processing resources may generate a data file storing annotations useful for downstream automated analysis, or downstream human analysis, such as timed bookmarks, rendered subtitles, among others. During downstream automated analysis, a machine learning model may utilize the annotations (e.g., the flagged data) as an input feature such that additional emphasis may be placed on annotated sections during the generation of output logits or normalized predictions.
In a variant embodiment, the modified capture of the video or audio data streams includes modifying an encoding approach (e.g., a compression approach) applied to the video or audio data streams prior to transmission to the centralized computer server. The raw video or audio data streams during the abnormality-related durations of time can be compressed differently such that a higher fidelity to the original data is maintained. This can be implemented through a modification of encoding characteristics, such as a level of lossy compression, the selective use of loss-less compression, the change in compression codec, a change in compression type (e.g., switching between or selectively applying discrete cosine transforms and motion compensation), increasing spatial or temporal redundancy, among others.
In this variant embodiment, the objective is to provide the centralized server with a greater set of features for analysis at the cost of increased downstream processing requirements and/or networking resources for transmission. To limit this potential increased demand on the system, limiting the use of or the requirement for the increased resource can be limited only to the durations of time where abnormalities are detected in the biometric data, and some durations of time temporally proximate to the abnormalities (e.g., +/−1 second, 2 seconds, 3 seconds, 4 seconds, 5 seconds).
In some embodiments, the biometric data is obtained via at least one of electrocardiography (ECG) data to assess the state of the autonomic nervous system and general cardiovascular data, electromyography (EMG) data to capture muscle fatigue, electroencephalography (EEG) to capture brain function, electrooculography (EOG) data and other methods to capture eye tracking, and/or “intelligent” shoe soles to assess the level of fatigue. In some embodiments, the alert may be tactile, visual and/or audible.
In some embodiments, the biometric data comprises ECG readings of a healthcare provider. The ECG readings may be converted into heart rate variability (HRV) values. When an HRV value passes below a threshold, the alert may be provided 2830 indicating that the healthcare provider is experiencing a high level of stress (e.g., acute stress) that is likely to cause an unwanted incident (e.g., due to performance of the healthcare provider) during the medical procedure. The healthcare provider may be relieved or perform coping actions to alleviate the stress. In some embodiments, the healthcare provider may return to the procedure once the HRV values of the near real-time biometric readings are back above the threshold value.
In some embodiments, the biometric data may be used to search for abnormalities in biometric readings in healthcare professionals. As noted above, the readings may indicate a high stress level. However, the readings may also indicate other issues. For example, stress level changes may be observed, including if stress levels are too low—i.e., if a healthcare provider is tired or lethargic, this could also lead to unwanted events and/or errors in the operating room. Therefore, receiving real-time or post-operative indications of all biometric abnormalities could be just as beneficial as receiving indications for high levels of stress.
In some embodiments, wearable technology and/or video analysis software may be used to capture biometric data (including, ECG, EMG, EEG, EOG, eye movement, muscle fatigue, other physiological measures, etc.), for the purposes of automating the detection of threats, adverse events, and other intraoperative segments of interest. In some embodiments, data attained via wearable technology and/or video analysis software may provide real time warnings and alerts (tactile, visual, or audible) to operating room teams. For example, during a live surgical procedure, biometric data capture may indicate that the lead surgeon has abnormally low HRV (an indicator of strong sympathetic nervous system activation, suggesting a highly stressed state). Such instances have been shown to be associated with adverse events, so knowledge of this via the proposed methods would be highly beneficial in the automation of the detection of adverse events.
Stress, in the context of some embodiments, is defined from data artifacts indicative of when the sympathetic nervous system dominating the parasympathetic nervous system are observed. Through experimentation, Applicants have considered various physiological markers that indicate this “dominance”, including lower HRV. When in a “fight-or-flight” mode, the time delta between consecutive heart beats, can, for example, become very consistent (in this example, the HRV is not measuring the heart rate, but rather the consistency of time deltas between beats). HRV metrics estimate the consistency of these time deltas.
For further illustration, when relaxed, time deltas may [700 ms, 900 ms, 800 ms, 1100 ms] show high variance, but when stressed they may [700 ms, 705 ms, 701 ms, 702 ms] show low variance. The time deltas derived from HRV can be a useful tool in controlling the allocation of processing, network, and recording resources in respect of an operating room black box system as described herein. This biometric information is useful for automatically focusing limited technical resources on regions of interest (e.g., durations of time) where abnormalities are detected.
In some embodiments, such data may be used for post-operative analytics to gain insight to the operating room environment in order to more effectively guide future practice. For example, a surgical resident may be shown to be highly sensitive to negative-feedback from an attending surgeon through biometric data capture, which contributes to this resident having significantly compromised surgical performance during such instances. Biometric data is crucial in determining such responses, and could be used for instructive purpose (e.g., suggesting stress management training for resident or recommending reduced negative feedback from surgeon to resident) in order to make the operating room environment as safe as possible for patients.
The process of using biometric data to identify and detect segments of interest (threats, adverse events, etc.) may assist with developing systems that increase awareness and thus potentially improving human performance and operating room safety. Such biometric data gives quantitative insights to the mental and physical state of the surgeon and other operating room staff that would otherwise be distracting, inefficient, inaccurate, and ultimate dangerous to communicate. For example, a team member under extreme stress (beyond ability to cope) due to some circumstances (operating room induced or otherwise) would not be able to accurately assess this stress level and attempts to do so would be distracting and thereby dangerous to the patient. A real time biometric assessment would be beneficial in such a scenario for both the purposes of bringing awareness to the stress level of the team member before compromised performance yields an adverse event (note, coping strategies for stress levels have been shown to be effective in reducing stress), in addition to gathering data for post-operative analysis (e.g., hypothetically, such data could be used to show that extreme stress states of operating room team members increase risk for surgical team mishaps and adverse events).
While capturing biometric data has typically been limited to cumbersome technology, wearable technology has evolved to make measurements of biometric data less obtrusive (e.g., ECG data can now be captured with simple devices like a watch or a t-shirt). Biometric data may be alternatively captured from video systems. For example, eye movement tracking is not limited to EOG measurements, as optical tracking software paired with a video camera can be used instead. Using wearable technology and video to capture biometric data is more feasible and less dangerous than using non-wearable technology to capture biometric data in the operating room.
In some embodiments, data attained via wearable technology and/or video analysis software may be synchronized with other data streams in the surgical operating room. For example, data acquired directly from this from this technology can be used in conjunction with other data sources from the Black Box platform (described further below) to determine relationships that may yield revelatory insights to guide future operating room standards and protocol.
An observational study was conducted at a hospital, collecting stress and surgical performance data were collected for one attending surgeon during a two month period. Physiological data to be used in calculations for proxy measures of acute stress were captured for the attending surgeon during each surgical procedure using the wireless Hexoskin Smart Shirt.
The Hexoskin Smart Shirt has built-in sensors that capture 256-Hz ECG data, three-dimensional 64-Hz acceleration data and 128-Hz breathing data. Data points were time-stamped to the Coordinated Universal Time (UTC). The 256-Hz ECG capture satisfies the recommended resolution to capture precise data to allow for HRV interpretations.
Consecutive elective surgical procedures, both laparoscopic and endoscopic, were included in the study. Each procedure was recorded and entirely anonymized of all markers (hospital information, physician identifiers, patient data). Video recordings were time-stamped to UTC to allow for synchronization with the HRV data captured by the Hexoskin Smart Shirt. Recordings were assessed independently and blindly by a single observer for technical laparoscopic surgical performance using the Operating Room Black Box platform, with a previously validated measurement tool.
In experimentation, both time domain and frequency domain approaches were considered, and it was noted that the two statistical methods correlate with one another to such a strong degree that they can be used as surrogates for each other. Further, time domain methods, particularly RMSSD (defined below), have been shown to have greater reliability when measuring short-term HRV than frequency domain methods.
As sympathetic nervous system activation begins to supersede parasympathetic nervous system activation, there becomes less interplay among these two ANS divisions, which manifests as a near-instantaneous decrease in RR-interval variation (lower HRV). As such, lower values for SDNN and RMSSD suggest greater levels of sympathetic nervous system dominance (hereafter referred to as stress). The Hexoskin ECG processing software provides a filter for data curation of RR intervals that excludes abnormal RR intervals. These may be due to irregular heartbeats or suboptimal ECG signals.
Technical laparoscopic surgical performance was evaluated with the GERT. The GERT allows for objective and reliable (interrater intraclass correlation coefficient for events 0.85) assessment of operative performance during laparoscopic procedures. Using this framework, both surgical errors and events are identified, with errors defined as ‘the mechanism of unintended or deviated technical task execution’ and events as ‘any deviation that causes injury to the patient or poses a risk of harm’. Examples of events include serosal tears arising from grasper slips from bowel, bleeding from needle puncturing vessel, and burns from inadvertent touching of other structures with an energy device.
Software was developed to derive the HRV statistics from the ECG data files, process the surgical performance data and synchronize all data streams into the pre-specified time intervals. The first interval in a procedure commenced the moment after the conclusion of the pre-surgery ‘time out’ checklist where video capture had begun. Data capture concluded at the end of the surgery when all surgical instruments were removed from the patient.
There is some debate regarding the optimal time intervals to ensure valid HRV recordings, with the Task Force of the European Society of Cardiology and the North American Society of Pacing Electrophysiology recommending that short-term HRV analysis be undertaken in intervals of 5 min, and others suggesting that accurate HRV recordings can be taken in intervals as low as 10 s. Consideration was also given to the fact that data streams were quite distinct. For example, although physiological data were measured on a pseudocontinuous timeline (ECG readings at 256 Hz), HRV was measured in discrete intervals and surgical performance on a continuous basis. Thus, in some embodiments, data streams are measured (HRV measurements and surgical performance) in non-overlapping intervals of 1, 2 and 5 min to best accommodate these differences. Stress data were divided into quantiles to allow for comparison between low and high stress intervals. Five-minute intervals were presented in quintiles, rather than tertiles, owing to the reduced sample variance resulting from the increased interval length, as stress peaks may last only a fraction of the duration of an interval.
Data was captured for gastric bypass (12 patients), sleeve gastrectomy (7) and peroral endoscopic myotomy (6), with a mean(s.d.) duration of 96·7(35·5) min for all procedures. As the rating framework for assessing surgical performance was tailored for laparoscopic procedures, surgical performance data were observed only for gastric bypass and sleeve gastrectomy. Stress data were observed for all procedures. Mean(s.d.) SDNN values were 38·0(14·4) ms for 1-min intervals, 40·9(12·2) ms for 2-min intervals and 43·7(9.7) ms for 5-min intervals. Respective mean(s.d.) RMSSD values were 25·2(16·5), 26·2(14·5) and 27·1(12·5) ms. The mean(s.d.) number of events per procedure was 23·8(11·3).
This study demonstrated a direct relationship between acute mental stress and surgical performance in a live operating theatre. The results indicate a clear association between the occurrence of events and the level of stress experienced, with significant differences between the higher and lower stress quantiles being observed for all measured time intervals. Greater stress responses were also found during intervals where an event was observed, for all measures except the 2-min interval RMSSD.
There are several possible explanations for this finding. One may be that increased stress levels in surgeons occur in response to the occurrence of events. Although this is likely to be true, it does not preclude the possible dependency of surgical performance on stress levels. Not all events are of equal severity and are therefore not necessarily recognized by the surgeon as a deviation that causes injury or poses a risk to the patient. Further, events that are recognized by the surgeon may not manifest in stress responses if they are perceived to be innocuous and easily resolved (for example, minor bleeding). In addition, the possibility of a snowball effect cannot be precluded, whereby an event triggers a stress response that could have a negative impact on surgical performance resulting in subsequent errors and events.
The true relationships may be more pronounced than has been found in the study. Stress responses vary greatly in length depending on the situation and, unless this stressed state fell perfectly within the bounds of an interval, the stressed state would either be diluted with more relaxed HRV data (if the stress response were shorter than the interval length) or be only partly captured (if longer than the interval length).
Observing at three different interval times was an attempt to control for this, but these methods could still be improved. In addition, an event occurring near the end of an interval would be unlikely to prompt an increased stress response during that same interval. Conversely, compromised surgical skill because of a raised stress response near the end of an interval may not trigger events until the succeeding interval. Similarly, if the event occurred at the very beginning of an interval, any preceding stress increase that may have been a factor would not be grouped into the same interval.
As described in further embodiments herein, the stress levels as measured as a proxy through biometric data can be utilized in conjunction with an operating theatre recording system to generate combined data that can be utilized to provide input data features into a machine learning model data architecture, such that one or more predictions could be established.
The clinical relevance of exploring variables associated with surgical events can be automatically tracked, as stress can be induced in surgeons and other members of the operating team for a variety of reasons, many of which are unavoidable. Machine learning approaches can be beneficial to distinguish between stress that manifests as a result of surgical events that demand increased concentration, and stress that arises as a result of external influences. The latter can be addressed by initiatives that identify and reduce such external events, whereas managing the former, which may or may not be harmful for surgical performance depending on the specific circumstance, is more complex.
In a first example, a machine learning system can be configured to predict a probability data output value indicative of an adverse event happening at time X in a particular procedure. Since HRV and adverse events are correlated, regardless of causality, HRV is can be a strong feature for predicting adverse events. Providing the processed biometric data adds a potential set of features that potentially increases the accuracy in predicting adverse events.
In a second example (unsupervised ML), the machine learning system can be configured for outlier detection. When biometric data deviates from the norm, it is possible that the deviation is due to some unusual circumstance that is a good target for analysis. Unsupervised learning approaches can be utilized for pattern analysis and prediction, for example, using self-organizing maps to assess probability densities over various inputs and to generate clusters for analysis. Conditional probability distributions can be established and, in some embodiments, principal component analysis may also be utilized to conduct exploratory data analysis.
In a third example, a machine learning data model can be configured to predict a probability that surgeon is experiencing burnout. HRV and other biometric markers can be utilized as a predictor for burnout and connected to personnel records (e.g., hours worked, hours of rest, scheduling) as input features to conduct pattern recognition.
Accordingly, in some embodiments, the biometric readings can be processed into derivative values, such as one or more heart rate variability (HRV) data values. These derivative values, if greater or lower than a pre-defined threshold data value, can then be processed to identify abnormality-related durations of time and tracked in the system in the form of time-based metadata tags. For example, HRV data values can be analyzed continuously for a trailing interval period of ECG values (e.g., 1 second, 2 seconds, 5 seconds, 10 seconds, 30 seconds, a minute, 2 minutes, 5 minutes, 30 minutes, 1 hour, 3 hours, 6 hours, 8 hours, 12 hours, 24 hours), and measured accordingly.
The pre-defined threshold data value, for example, can be associated with different measures, such as SDNN and RMSSD operating as proxy values for stress (e.g., sympathetic nervous system dominance). Threshold data values can be person-dependent in some embodiments, and accordingly, a baseline could be set and the thresholds may be established based on standard deviation ranges relative to baseline.
BlackBox System/Platform
Embodiments described herein may provide device, system, method, platform and/or computer readable medium which provides comprehensive data collection of all details of patient care in one or more such settings to: identify and/or analyze errors, adverse events and/or adverse outcomes; provide comprehensive data allowing investigation of the chain of events from an error to adverse events; provide information concerning individual and/or team performance, e.g., for high-stakes assessment of competence, certification and/or re-certification of healthcare professionals; provide data to be used for design of individualized training interventions for surgical and/or medical teams based on demonstrated performance deficiencies; identify critical safety deficiencies in human performance and/or safety processes, e.g., for creation of individualized solutions aimed to reduce risks and/or enhance patient safety; and/or assess critical safety deficiencies in medical technology and/or provide feedback for improvement in design and/or performance, analyze and monitor efficiency and safety processes in a clinical environment.
The system is augmented with an additional data stream that is provided through a biometric sensor coupled to a body of a healthcare practitioner. The biometric sensor can include a sensor device coupled to a shirt, embedded into shoes, coupled into an armband, incorporated into a smart watch or a smart phone, among others. There may be one or more biometric sensors. An example biometric sensor array includes a Hexoskin smart shirt, which is a clinically validated shirt adapted for continuous ECG, HRV, RRV determinations, among others.
The system can include a local computer server that resides in or is coupled directly to the operating theatre, which is configured for controlling recording activities and initial encoding prior to transmission to a central processing engine, which can be coupled across a network to one or more local computer servers and tasked with conducting more intensive data processing and machine learning modelling based on data received from the local computer servers.
In an aspect, embodiments described herein relate to a system for collecting and processing medical or surgical data. The system may have a plurality of hardware units for collecting real-time medical or surgical data streams having a control interface coupled by a network to cameras, sensors, audio devices, biometric sensing devices, and patient monitoring hardware, the real-time medical or surgical data streams relating to a real-time medical procedure within an operating or clinical site.
The hardware units may gather or collect one or more independent data streams from different devices, and in turn each data stream provided the hardware unit may be independent of other data streams provided by other hardware units. According, the system may implement synchronization techniques of the data streams as described herein. The system may have device middleware and hardware for translating, connecting, and formatting the real-time medical or surgical data streams received independently from the hardware units (which in turn may receive data feeds from different devices independently).
As described herein, the hardware units have can have modifiable recording characteristics that are controlled or coordinated by a local computer server, or in some instances, a coupled cloud based or central computing server. The local computer server includes one or more computer processors, and is coupled to the biometric devices to obtain data stream recordings of biometric data.
The biometric data can then be processed locally for establishing time synchronization with audio/video feeds. A technical challenge with surgical recordings results from a lack of synchronization of audio/video feeds with digital data streams (e.g., from sensors). Synchronization is an important feature that may be significantly advantageous in converting unstructured data into structured data, for example, as inputs into a perception engine (e.g., that can utilized automate the assessment process and/or provide other types of machine-learning outputs, predictions and/or estimations). For example, replacing (and/or reducing) a need for manual evaluation of the recordings with automated software may help provide a cheap and scalable approach.
The biometric data can be captured at different levels of representation, according to various embodiments. For example, the data can be raw data, such as ECG data values at particular timestamps, showing a voltage versus time of electrical activity of the heart. These can be obtained, for example, via electrodes positioned on limbs or a chest, such as in a specially configured sensor-loaded shirt. In another embodiment, the ECG data values are then processed to extract one or more derivative feature sets, such as P-wave features, QRS complex features, T wave features, and readings thereof, such as amplitudes, frequencies, based on graphical deflections corresponding to depolarizations of various chambers of the heart through one or more heart beats.
The ECG data can, in an embodiment, be utilized to establish HRV measures as described herein, which can be adapted as a proxy of a fight-or-flight response. Experimentation indicates that the HRV measure, in particular, is a useful mechanism to provide an automated approach for measuring a proxy value for stress.
The system may have an encoder with a network server for synchronizing and recording the real-time biometric, medical or surgical data streams to a common clock or timeline to generate a session container file. As noted, the synchronization may aggregate independent data feeds in a consistent manner to generate a comprehensive data feed generated by data from multiple independent devices.
The system may have network infrastructure connecting the encoder, the device middleware and hardware and the hardware units, and switching or gateway hardware for a virtual private network to transmit the session container file. The session container file, for example, can contain an additional stream of data, which, in some embodiments, can include a raw biometric data stream, or a set of generated time-based metadata tags indicative of the one or more abnormality-related durations of time that are encapsulated into the output data structure.
As described herein, the ECG data can then be utilized to modify characteristics of the operation of the encoder, modifying recording, encoding, and/or transmission characteristics of the encoder relating to the session container file. In a specific embodiment, abnormalities found in the HRV data (e.g., flagging durations of time where HRV is one, two, or three standard deviations away from a baseline) can be utilized to establish a series of metadata tags that can be generated and/or appended into the session container file to represent flagged periods of time.
During the flagged periods of time, the encoder operation can be modified, as these flagged periods of time may be related to an adverse event starting to occur or in occurrence. In a first example, the encoder operation modifies capture characteristics of the other data streams (e.g., audio and video), by increasing a resolution and/or bitrate of capture. In this example, a normal bitrate of 5 Mbps (corresponding to 720p video) is shifted up to 35-45 Mbps (corresponding to 4K video) for the duration. Audio bitrate can be increased, for example, from a 128 kilobits per second recording to a 320 kilobits per second recording.
In another embodiment, the flagged periods of time are used by the local encoder to actuate additional sensors to provide enhanced streams. For example, additional cameras, microphones, and other sensory devices may be only switched on or recorded during these periods of tracked abnormalities. The additional cameras, for example, can include an additional overhead camera with a wide angle of capture that may not otherwise be utilized as the wide angle of capture yields a large amount of normally extraneously data (e.g., capturing movements outside of the surgical focus area), which in a normal operating context may not be particularly valuable to a downstream analysis. However, in situations where an abnormality is detected, the wide angle of capture can be useful as it may capture activities that are taking place by other practitioners to rectify a problem, and can be used for future predictions of adverse events (e.g., a recording captured a nurse practitioner moving quickly to address an unexpected bleeding wound, and in these situations, adverse events were avoided).
The increased size of the audio/video causes the data streams to be more computationally intensive to record, process and store. By limiting the enhanced encoder operation only to the flagged periods of time, the overall burden is reduced while providing an enhanced set of features for downstream analysis and processing, for example, by the central computing server or a cloud based implementation.
In some example embodiments, the device middleware and hardware establishes a secure reliable connection using the network infrastructure for communication with the encoder and the hardware units. During the abnormality related durations of time, the network infrastructure automatically allocates priority communication channels and/or bandwidth for the increased load during the flagged periods of time. In particular, a network interface handler receives a control signal, such as data packet indicative of a control command or an analog voltage up signal indicating that increased data communications will be required for a short period of time.
The control signal may last, for example, for the entire duration of an abnormality related duration of time, and in some embodiments, may be adapted to have an additional trailing period of time soon after (e.g., 1 second, 5 seconds, 30 seconds) to account for any encoding lag. For example, a larger proportion, or all of a gigabit Ethernet connection (e.g., 1000BASE-T) can be allocated the transmission, taking priority over other transmissions (which may then be lagged as they await communication resources to become free again), such as routine data alerts, machine heartbeat signals, among others.
In this example, a larger communications pipe may be established to provide for a large burst of communications. The priority communication channels can, in some embodiments, be utilized for increased communication to a backend perception engine as described herein to conduct prioritized machine learning to provide real or near-real time analysis of the encoded audio, video, and/or biometric data.
In some example embodiments, the device middleware and hardware implements data conformity and accurate synchronization for the real-time medical or surgical data streams using network protocols for clock synchronization between the hardware units to assist the encoder to generate the session container file.
The perception engine, for example, may be utilized to provide a multi-nodal approach for machine learning. The perception engine may receive as inputs, for example, various recording data, physiological data, biometric data, among others. The perception engine can be a centralized computer server or a cloud based computer server having increased computational resources (e.g., significant multi-threaded processor capabilities and graphics processing capabilities, a large amount of RAM (such as 32 GB of RAM), significant high-speed storage devices). The centralized computer server or a cloud based computer server, in a further embodiment is configured for a capability to request additional computational resources from a set of available distributed resources, for example, provisioned by “spinning up” or deprovisioned by “spinning down” various coupled computing devices.
The perception engine may be configured for generating a protocol for data extraction from the session container file, processing the data using the protocol to extract patterns for time-stamped clinical events within the session container file, generating an interface indicator for a temporal sequence of the time-stamped clinical events within the session container file and error assessments, and generating a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures with neural networks for modelling correlation of data for interference and feature extraction.
The perception engine then processes the session container files to conduct machine learning analyses. As described herein, a technical challenge is the limited availability of computing resources. Data processing power, especially in relation to machine vision and machine audio tasks, is very computationally expensive. For example, every single pixel of every single frame can be represented by 24 bits in RBG). This causes technical limitations (speed, compute resources, labor for labelling training data, etc.). While ECG data is comparatively smaller, it still provides an increase in complexity in analysis as the number of input features increase as independent variables in a feature space expand.
As described herein, a limited approach to requesting resources can be established through only activating increased input feature analysis during the durations of time that are flagged in the metadata as having an abnormality in biometric data. The improved encoded audio and video during this duration of time provides improved predictive power, at the cost of computational complexity. In some embodiments, the local computer server or the centralized computer server are configured to request or allocate additional computational resources in anticipation of the need to generate real-time or near-real time alerts. For example, such an approach is helpful. Coupled with biometric analysis, this approach limits the amount of technical computational resources required while helping automatically focus the usage of resources during periods of time in which the additional predictive power is particularly useful and actionable. The use of HRV, in particular, as a proxy for stress was experimentally found to be a useful predictor of potential adverse events.
In a first embodiment, alerts are controlled to be actuated during the entire duration of the flagged abnormality (e.g., an alert light may be turned on, a flashing signal may be controlled, a graphical display may occur, a tactile alert such as a vibration could occur). However, this naïve approach may yield a large number of false positive events, causing alert fatigue in practitioners. False positives can occur, for example, in particularly stressful procedures (e.g., a less often conducted surgical procedure or a procedure with a high complication rate) as practitioners may naturally have HRV values that indicate stress levels that pass a threshold for a baseline and there may be many durations of abnormality relative to a low stress procedure, such as a routine gallbladder removal. On the other hand, a high measured proxy for stress for gallbladder removal may be more concerning.
In a second embodiment, alerts are controlled to be actuated only when the perception engine generates an output predictive of an increased probability of an adverse event occurring greater than a particular threshold. In this embodiment, the perception engine receives increased input features for conducting an analysis (and the flag and/or the underlying biometric data itself may be an input), the perception engine's capability for generating accurate predictions is increased but automatic tuning can occur through training the perception engine over a number of epochs of training cases, or continuous training.
The alerts can be controlled in this situation to be more closely aligned with the predicted probability of adverse event, reducing potential false positives, for example, associated with a less often conducted surgical procedure, as an abnormal biometric reading is not always predictive of an adverse event occurring. In some embodiments, the perception engine is configured for secondary event tracking, such as predicting events that could lead to morbidity in a post-operative time period.
For example, the obtained data may be provided in the form of audio clippings recorded from surgical procedures, and the perception engine may apply machine learning techniques to generate automated predictions. These predictions may be verified and/or compared against records of tracked incidents and/or events for accuracy, and the perception engine may be tuned over a period of time based on the particular outputs desired, their accuracy, specificity, and sensitivity, among others. The data extraction may involve extracting audio data for processing. The extraction may involve extracting video data for processing.
The machine-learning approaches may include alternating decision trees, random forests, among others, and may be tuned for application in relation to surgical or medical contexts. Where the perception engine is being used for analysis of medical and/or clinical events (e.g., a surgical error/adverse outcome), the perception engine may be tuned to improve the likelihood of detecting true positives, at the cost of increasing the likelihood of detecting false positives. For example, the perception engine may be utilized to provide instruction sets such as metadata indicative of when a medical reviewer should review a surgical procedure. In such a scenario, it may be advantageous for the perception engine to apply a broader scope of detection, possibly including false positives (which can then be verified and discarded by a medical reviewer).
The outputs from the perception engine may be provided in the form of various interfaces, such as graphical user interfaces for indication of when clinical events have been estimated, or application programming interfaces for computerized interactions with third party devices and/or databases.
In some example embodiments, the encoder and device middleware and hardware are operable to interface with third party devices to receive additional data feeds as part of the real-time medical or surgical data streams.
In some example embodiments, the system has a central control station accessible using the control interface, the control station configured to control processing of the data streams in response to input control comprising play/pause, stop session, record session, move to session frame, split-display, recording status indicator, and log file.
In some example embodiments, the network infrastructure provides increased fail-over and redundancy for the real-time medical or surgical data streams from the hardware units.
In some example embodiments, the system has a storage area network for storing data container files of the real-time medical or surgical data streams until scheduled transmission.
In some example embodiments, the encoder implements identity anonymization and encryption to the medical or surgical data.
In some example embodiments, the encoder processes the real-time medical or surgical data streams to generate measurement metrics relating to the medical procedure.
In some example embodiments, the real-time medical or surgical data streams correlates to a timeline, wherein the encoder detects events within the real-time medical or surgical data streams at corresponding times on the timeline, and tags and timestamps the session container file with the events, the timestamps corresponding to times on the timeline.
In some example embodiments, the system has an intelligent dashboard interface for annotation and tagging of the synchronized medical or surgical data streams, wherein the intelligent dashboard may implement a viewer with playback viewing for reviewing content and interface controls for tagging content.
In some example embodiments, the intelligent dashboard is multi-dimensional in that the union of all dimension variables for the medical procedure as represented by the real-time medical or surgical data streams may indicate a specific set of one or more applicable annotation dictionaries or coding templates.
In some example embodiments, example variables that may be used to determine the annotation and tagging dictionary may be: the type of medical procedure being performed, the aspect of the procedure that is being analyzed, the geographic area/region where the procedure is being performed.
In another aspect, there is provided a multi-channel encoder for collecting, integrating, synchronizing and recording medical or surgical data streams onto a single interface with a common timeline or clock, the medical or surgical data streams received as independent real-time or live data streams from a plurality of hardware units, the encoder having a network server for scheduling transmission of session file containers for the recordings, the encoder processing the medical or surgical data streams to generate measurement metrics relating to a real-time medical procedure. The encoder aggregates multiple independent data streams or feeds received from different hardware unit and smart devices.
In some example embodiments, the encoder generates as output a single session transport file using lossless compression operations.
In some example embodiments, the encoder detects completion of a recording of the data streams and securely encrypts the single transport file.
In some example embodiments, the encoder implements identity anonymization to the medical or surgical data.
In some example embodiments, the data streams include audio, video, text, metadata, quantitative, semi-quantitative, and data feeds.
In another aspect, there is provided a method for collecting and processing medical or surgical data. The method involves receiving, at a multi-channel encoder, a plurality of live or real-time independent input feeds from one or more data capture devices located in an operating room or other patient intervention area, the input feeds relating to a live or real-time medical procedure;
The method may involve synchronizing, by the encoder, the plurality of live independent input feeds onto a single interface with a common timeline or clock, and recording the synchronized input feeds using a network server. The method may involve generating, by the encoder, an output session file using the synchronized input feeds, and transmitting the output session file using the network server.
In some example embodiments, the method further involves processing the data streams for identity anonymization.
In some example embodiments, the method further involves routing the data streams using a switch router to the encoder.
In a further aspect, there is provided a cloud based system for collecting and processing medical or surgical data. The system may have an encoder having a control interface for, in response to receiving a control command, triggering collection of real-time medical or surgical data streams by smart devices including cameras, sensors, audio devices, and patient monitoring hardware, the medical or surgical data relating to a real-time medical procedure within an operating or clinical site, the encoder for authenticating the smart devices, the smart devices synchronizing the real-time medical or surgical data streams by embedding timestamp markers within the real-time medical or surgical data streams, the timestamp markers generated by each smart device by a device clock. The system also has a media management hub server with middleware and hardware for translating, connecting, formatting, and recording the real-time medical or surgical data streams to generate session container files on network accessible storage devices, and wireless network infrastructure to provide a secure network connection between the encoder, the smart devices and the media management hub server for communication of the real-time medical or surgical data streams. The system has a central content server for storing and distributing the session container files and providing a two-way communication interface for the media management hub to implement a file transfer handshake for the session container files. The system has switching or gateway hardware for a virtual private network tunnel to transmit the session container files from the media management hub to the central content server. The cloud based system may enable autonomous, independent smart devices to time stamp collected data and implement synchronization techniques to aggregate independent data streams and feeds to generate a comprehensive, real-time data representation of the medical or surgical procedure or unit.
In some example embodiments, the media management hub server broadcasts clock data to the smart devices for synchronization of the device clocks.
In some example embodiments, the encoder provides a user interface to receive the control command and display real-time visual representations of the medical or surgical data.
In some example embodiments, the media management hub server aggregates, packages, compresses and encrypts the real-time data streams to generate the session container files.
In some example embodiments, the media management hub server manages the smart devices based on location, schedule, zone and requirements.
In some example embodiments, the media management hub server receives operating status data from the smart devices to generate a management interface with a visual representation of the operating status data for the smart devices, the operating status data including online, offline, running capture, and on-board storage.
In some example embodiments, the media management hub server processes the operating status data to detect smart devices operating outside of normal conditions and in response generating an alert notification of the detected smart devices operating outside of normal conditions.
In some example embodiments, the media management hub server implements a device communication interface for the smart devices to implement a device data transfer handshake for the real-time medical or surgical data streams.
In some example embodiments, the media management hub server authenticates the smart devices.
In some example embodiments, the system has a computational intelligence platform for receiving the session container files to construct an analytics model to identify clinical factors within the session container files for predictions, costs and safety hazards, the analytics model providing a network for extracting features, correlations and event behaviour from the session container files that involve multivariable data sets with time-variant parameters.
In some example embodiments, the system has a training or education server to receive the session container files, process the session container files to identify root causes of adverse patient outcomes and generate a training interface to communicate training or performance feedback data using the identified root causes and the session container files.
In some example embodiments, the smart devices include motion tracking devices for markerless motion tracking of objects within the operating or clinical site, the system further comprising a processor configured to convert captured motion data from the motion tracking devices into data structures identifying human factors, workflow design and chain-of-events.
The platform may have different aspects including hardware, software, front end components, middleware components, back end components, rich content analysis software and analytics software (e.g., database).
Multiple processors may be configured with operating system and client software (e.g., Linux, Unix, Windows Server, or equivalent), scheduling software, backup software. Data storage devices may be connected on a storage area network.
The platform 10 includes hardware units 20 that include a collection or group of data capture devices for capturing and generating medical or surgical data feeds for provision to encoder 22. The hardware units 20 may include cameras 30 (e.g., wide angle, high definition, pan and zoom camera, such as a Sony EVI-HD1™ or other example camera) mounted within the surgical unit, ICU, emergency unit or clinical intervention units to capture video representations of the OR as video feeds for provision to encoder 22. The video feed may be referred to as medical or surgical data. An example camera 30 is a laparoscopic or procedural view camera (AIDA™, Karl Storz™ or equivalent) resident in the surgical unit, ICU, emergency unit or clinical intervention units. Example video hardware includes a distribution amplifier for signal splitting of Laparoscopic cameras. In some embodiments, not all cameras 30 are utilized at all times, and rather, selective operation of the cameras 30 may be controlled to limit an overall communication/data size burden associated with the recordings. The hardware units 20 can have controllable aspects in relation to how recordings are captured, including, for example, the ability to change resolutions, bitrates, encoding patterns, encoding codecs (e.g., selectively applying motion interpolation), among others. The hardware units 20 can be controllable responsive to certain local tracked data objects, such as abnormalities detected in a stream of biometric capture data from a biometric device coupled to a practitioner.
The hardware units 20 have audio devices 32 (e.g., condenser gooseneck microphones such as ES935ML6™, Audio Technica™ or other example) mounted within the surgical unit, ICU, emergency unit or clinical intervention units to provide audio feeds as another example of medical or surgical data. Example sensors 34 installed or utilized in a surgical unit, ICU, emergency unit or clinical intervention units include but not limited to: environmental sensors (e.g., temperature, moisture, humidity, etc., acoustic sensors (e.g., ambient noise, decibel), electrical sensors (e.g., hall, magnetic, current, mems, capacitive, resistance), flow sensors (e.g., air, fluid, gas) angle/positional/displacement sensors (e.g., gyroscopes, altitude indicator, piezoelectric, photoelectric), and other sensor types (e.g., strain, level sensors, load cells, motion, pressure). The sensors 34 provide sensor data as another example of medical or surgical data. The hardware units 20 also include patient monitoring devices 36 and an instrument lot 18.
The customizable control interface 14 and GUI (may include tablet devices, PDA's, hybrid devices, convertibles, etc.) may be used to control configuration for hardware components of unit 20. The platform 10 has middleware and hardware for device-to-device translation and connection and synchronization on a private VLAN or other network. The computing device may be configured with anonymization software, data encryption software, lossless video and data compression software, voice distortion software, transcription software. The network hardware may include cables such as Ethernet, RJ45, optical fiber, SDI, HDMI, coaxial, DVI, component audio, component video, and so on to support wired connectivity between components. The network hardware may also have wireless base stations to support wireless connectivity between components.
Descriptions and Definitions for an Illustrative Embodiment
Illustrative definitions of various components are provided as examples of various embodiments.
A Private VLAN may refer to a networking technique, which provides network segregation and secure hosting of a network on the clients, existing backbone architecture via restricted “private ports”.
A VPN may extend a private network across a public network, such as the Internet. It enables a computer or network-enabled device to send and receive data across shared or public networks as if it were directly connected to the private network, while benefiting from the functionality, security and management policies of the private network.
Anonymization Software for anonymizing and protecting the identity of all medical professionals, patients, distinguishing objects or features in a medical, clinical or emergency unit. This software implements methods and techniques to detect facial, distinguishing objects, or features in a medical, clinical or emergency unit and distort/blur the image of the distinguishing element. The extent of the distortion/blur is limited to a localized area, frame by frame, to the point where identity is protected without limiting the quality of the analytics.
Voice or Vocabulary Alteration Software for anonymizing and protecting the identity of all medical professionals, patients, distinguishing objects or features in a medical, clinical or emergency environment. This software may implement methods and techniques running on hardware in a medical, clinical or emergency environment to alter voices, conversations and/or remove statements of everyday language to preserve the identity of the speaker while at the same time maintaining the integrity of the input stream so as to not adversely impact the quality of the analytics.
Data Encryption Software may execute to encrypt computer data in such a way that it cannot be recovered without access to the key. The content may be encrypted at source as individual streams of data or encrypted as a comprehensive container file for purposes of storage on an electronic medium (i.e. computer, storage system, electronic device) and/or transmission over internet 26. Encrypt/decrypt keys may either be embedded in the container file and accessible through a master key, or transmitted separately.
Lossless Video and Data Compression software executes with a class of data compression techniques that allows the original data to be perfectly or near perfectly reconstructed from the compressed data.
Device middleware and hardware may be provided for translating, connecting, formatting and synchronizing of independent digital data streams from source devices. The platform 10 may include hardware, software, algorithms and methods for the purpose of establishing a secure and reliable connection and communication directly, or indirectly (via router, wireless base station), with the OR encoder 22, and third-party devices (open or proprietary) used in a surgical unit, ICU, emergency or other clinical intervention unit.
The hardware and middleware may assure data conformity, formatting and accurate synchronization. Synchronization may be attained by utilizing networking protocols for clock synchronization between computer systems and electronics devices over packet-switched networks like NTP, etc.
The hardware unit may include third party devices (open or proprietary) non limiting examples being O2 saturation monitors, anesthesia monitors, patient monitors, energy devices, intelligent surgical devices (i.e. smart staplers, smart laparoscopic instruments), autonomous surgical robots, etc. hospital patient administration systems (i.e. electronic patient records), Intelligent implants, Sensors including but not limited to: Environmental sensors: i.e. temperature, moisture, humidity, etc. Acoustic sensors: i.e. ambient noise, decibel, etc. Electrical sensors: i.e. hall, magnetic, current, mems, capacitive, resistance, etc. Flow sensors: i.e. air, fluid, gas, etc. angle/positional/displacement sensors: i.e. gyroscopes, attitude indicator, piezoelectric, photoelectric, etc. Other sensors: strain, level sensors, load cells, motion, pressure, and so on.
Transcription Software may assist in the conversion of human speech into a text transcript utilizing technologies such as natural language speech recognition.
OR or Surgical encoder: The OR or Surgical encoder (e.g., encoder 22) may be a multi-channel encoding device that records, integrates, ingests and/or synchronizes independent streams of audio, video, and digital data (quantitative, semi-quantitative, and qualitative data feeds) into a single digital container. The digital data may be ingested into the encoder as streams of metadata and is sourced from an array of potential sensor types and third-party devices (open or proprietary) that are used in surgical, ICU, emergency or other clinical intervention units. These sensors and devices may be connected through middleware and/or hardware devices which may act to translate, format and/or synchronize live streams of data from respected sources.
The encoder 22 is configured, in some embodiments, to modify operational characteristics of the capture in response to one or more biometric data stream-based determinations, such as a determination that an abnormality is detected for a particular duration of time (e.g., out of bounds HRV value that is beyond 1-3 standard deviations from a baseline). The operational characteristics of the capture can include temporary modifications by increasing resolution, bitrate, reducing compression factors (to increase fidelity of an output), actuating additional recording devices, requesting additional bandwidth resources, or requesting additional downstream processing resources.
The modifications to the operation of the devices can be actuated through generating command signals to change specific operational parameters. The modifications to request additional bandwidth may include modifying network interface operating characteristics, such as increasing allocations or decreasing allocations of specific communication channels or frequencies. The modifications to request additional processing resources may include selectively requesting additional resources to be provisioned so that a downstream perception engine is able to more readily generate real or near-real predictions to provide an improved machine learning based alert system.
Customizable Control Interface and GUI. The Control Interface (e.g., 14) may include a Central control station (non-limiting examples being one or more computers, tablets, PDA's, hybrids, and/or convertibles, etc.) which may be located in the clinical unit or another customer designated location. The Customizable Control Interface and GUI may contain a customizable graphical user interface (GUI) that provides a simple, user friendly and functional control of the system.
Example features of the Customizable Control Interface and GUI may include but are not limited to: Play/Pause button which may enable some segments of the procedure to not be recorded. To omit these segments from the recording, the user interface can pause the recordings and re-start when desired. The pause and play time-stamps are recorded in a log file, indicating the exact times of the procedure that were extracted; a stop session button can be provided that when selected, files are closed and automatically transferred to the storage area network (SAN); split-screen quadrant display of video feeds can be provided, which may provide visual displays of videos in real-time during recording; Visual indicator of recording may be a colored, blinking dot appeared on screen to provide visual indication to the team that video and audio feeds are being recorded; Log file where at the end of the recording, a log file may be generated that indicates key time points, including start and end time of the recording session, pauses and replays; Password protection, which may refer to an interface that is secured with one or several layers of password protection to ensure maintenance of patient confidentiality and privacy.
System Level Application may refer to a platform 10 that is designed to be a scalable platform ranging from small single clinical intervention unit to large-scale clinical intervention unit(s). Where necessary, a switching router may be used in larger scale applications to maximize efficiency and/or deliver increased fail-over and redundancy capabilities.
Example Applications
In an aspect, embodiments described may provide an illustrative small scale application. As a small single encoder platform, audio, video and data feeds are connected to the encoder 22 directly via cable or indirectly via connected wireless base station.
Using the Customizable Control Interface and GUI, activation of the system may commence recording, collection and streaming of all available audio, video, sensor and data feeds (which may be referred to as medical and surgical data feeds) to the encoder 22. It will use all available cameras including both mounted and laparoscopic, all audio microphones and all available and implemented sensors and third-party devices (open or proprietary) used in a surgical units, ICU, emergency or other clinical intervention units. Pause or Stop or Play commands will send corresponding commands to the encoder 22. Digital data is formatted, translated and synchronized through middleware hardware and software and using networking protocols for clock synchronization across the network. Digital data can be ingested into the encoder 22 as metadata.
The encoder 22 may be responsible for synchronizing all feeds, encoding them into a signal transport file using lossless audio/video/data compression software.
Upon completion of the recording, the container file will be securely encrypted. Encrypt/decrypt keys may either be embedded in the container file and accessible through a master key, or transmitted separately. The encrypted file may either be stored on the encoder 22 or stored on a storage area network until scheduled transmission.
The communications server on the private VLAN will be responsible for schedule management and the automated file and key transmission. This may be done through a private VLAN on the client environment and transmitted via Virtual Private Network (VPN) 24 on public data lines directed back to a back office.
The communications server may be responsible for backing up data including audio, video, data, encrypted files, etc. utilizing backup software as part of the configuration.
The communications server may be responsible for hosting and directing all traffic between the private VLAN and back office.
In another aspect, embodiments described herein may involve an encoder configured for hosting and operating anonymization and voice or vocabulary alteration software(s) for the purpose of protecting the identity of medical professionals, patients, distinguishing objects or features in a medical, clinical or emergency environment. This may be done either before compressing, containerizing and/or encrypting the collective data, or after receipt of transmission to back office and decryption.
In an aspect, embodiments described may provide an illustrative larger scale application.
Larger application environments may be required. In order to maximize efficiency and deliver increased fail-over and redundancy capabilities, a switching router may be used (e.g., router 16 of
Using the Customizable Control Interface 14 and GUI, activation signals may trigger or commence recording, collection and streaming of all available audio, video and data feeds (from components of hardware units 20) to one of multiple available encoders 22 via the switching router 16. For example, the data stream or feeds may be from all available cameras including both mounted and laparoscopic, all audio microphones and all available and implemented sensors and third-party devices (open or proprietary) used in hardware units 20 which may relate to surgical units, ICU, emergency or other clinical intervention units. Control commands such as Pause/Stop/Play commands received at Control Interface 14 may send corresponding control commands to the encoder 22. Digital data may be formatted, translated and synchronized through middleware hardware and software and using networking protocols for clock synchronization across the network. Digital data streams may be ingested into the encoder 22 as Metadata. The encoder 22 may be responsible for synchronizing all feeds and encoding them into a signal transport file using lossless audio/video/data compression software.
Upon completion of the recording, the container file may be securely encrypted. Encrypt/decrypt keys may either be embedded in the master file and accessible through a master key, or have a separate key. The encrypted file will either be stored on the encoder 22 or stored on a Storage area network until scheduled transmission.
The communications server on the private VLAN 24 may be responsible for schedule management and the automated file and key transmission. This may be done through a private VLAN on the client environment and transmitted via VPN 24 on public data lines directed back to a back end office, or other system.
The communications server (e.g., network server 12) may be responsible for backing up data including audio, video, data, encrypted files, etc. utilizing backup software as part of the configuration. The communications server may be responsible for hosting and directing all traffic between the private VLAN and back office system, for example.
In some examples, encoder 22 may also be responsible for hosting and operating anonymization and voice/vocabulary distortion software(s) for the purpose of protecting the identity of all medical professionals, patients, distinguishing objects or features in a medical, clinical or emergency environment captured in data streams of hardware units 20. This may be done either before compression, containerizing and encryption, or after decrypting in back office system.
In an aspect, embodiments described herein may provide a device, system, method, platform and/or computer readable medium which is housed in clinical areas and allows gathering of comprehensive information from every aspect of the individual, team and/or technology performances and their interaction during clinical interventions. The data capture devices may be grouped as one or more hardware units 20 as shown in
According to some embodiments, this information may include: video from the procedural field; video of the clinical environment; audio; physiological data from the patient; environmental factors through various sensors (e.g., environmental, acoustic, electrical, flow, angle/positional/displacement and other potential sensors); software data from the medical devices used during intervention; and/or individual data from the healthcare providers (e.g., heart rate, blood pressure, skin conductance, motion and eye tracking, etc.). In some embodiments, the a heart rate variability of the healthcare provider may be used to determine acute stress on the part of the healthcare provider.
According to some embodiments, this information then may be synchronized (e.g., by the encoder 22) and/or used to evaluate: technical performance of the healthcare providers; non-technical performance of the clinical team members; patient safety (through number of registered errors and/or adverse events); occupational safety; workflow; visual and/or noise distractions; and/or interaction between medical/surgical devices and/or healthcare professionals, etc.
According to some embodiments, this may be achieved by using objective structured assessment tools and questionnaires and/or by retrieving one or more continuous data streams from sensors 34, audio devices 32, an anesthesia device, medical/surgical devices, implants, hospital patient administrative systems (electronic patient records), or other data capture devices of hardware unit 20.
According to some embodiments, significant “events” may be detected, tagged, time-stamped and/or recorded as a time-point on a timeline that represents the entire duration of the procedure and/or clinical encounter. The timeline may overlay captured and processed data to tag the data with the time-points.
Upon completion of data processing and analysis, one or more such events (and potentially all events) may be viewed on a single timeline represented in a GUI, for example, to allow an assessor to: (i) identify event clusters; (ii) analyze correlations between two or more registered parameters (and potentially between all of the registered parameters); (iii) identify underlying factors and/or patterns of events that led to adverse outcome; (iv) develop predictive models for one or more key steps of an intervention (which may be referred to herein as “hazard zones”) that may be statistically correlated to error/adverse event/adverse outcomes, v) identify a relationship between performance outcomes and clinical costs. These are non-limiting examples of uses an assessor may make of a timeline presented by the GUI representing recorded events.
Analyzing these underlying factors according to some embodiments may allow one or more of: (i) proactive monitoring of clinical performance; and/or (ii) monitoring of performance of healthcare technology/devices (iii) creation of educational interventions—e.g., individualized structured feedback (or coaching), simulation-based crisis scenarios, virtual-reality training programs, curricula for certification/re-certification of healthcare practitioners and institutions; and/or identify safety/performance deficiencies of medical/surgical devices and develop recommendations for improvement and/or design of “intelligent” devices and implants—to curb the rate of risk factors in future procedures and/or ultimately to improve patient safety outcomes and clinical costs.
The device, system, method and computer readable medium according to some embodiments, may combine capture and synchronization, and secure transport of video/audio/metadata with rigorous data analysis to achieve/demonstrate certain values. The device, system, method and computer readable medium according to some embodiments may combine multiple inputs, enabling recreation of a full picture of what takes place in a clinical area, in a synchronized manner, enabling analysis and/or correlation of these factors (between factors and with external outcome parameters (clinical and economical). The system may bring together analysis tools and/or processes and using this approach for one or more purposes, examples of which are provided herein.
Beyond development of a data platform 10, some embodiments may also include comprehensive data collection and/or analysis techniques that evaluate multiple aspects of any procedure. One or more aspects of embodiments may include recording and analysis of video, audio and metadata feeds in a synchronized fashion.
The data platform 10 may be a modular system and not limited in terms of data feeds—any measurable parameter in the OR/patient intervention areas (e.g., data captured by various environmental acoustic, electrical, flow, angle/positional/displacement and other sensors, wearable technology video/data stream, etc.) may be added to the data platform 10. One or more aspects of embodiments may include analyzing data using validated rating tools which may look at different aspects of a clinical intervention. These aspects may include: technical performance, non-technical “team” performance, human factors, patient safety, occupational safety, workflow, audio/visual distractions, etc.
Video, audio and synchronized metadata may be analyzed using manual and/or automatic data analysis techniques, which may detect pre-determined “events” that can be tagged and/or time-stamped. All tagged events may be recorded on a master timeline that represents the entire duration of the procedure. Statistical models may be used to identify and/or analyze patterns in the tagged events. Various embodiments may encompass a variety of such statistical models, current and future.
According to some embodiments, all video feeds and audio feeds may be recorded and synchronized for an entire medical procedure. Without video, audio and data feeds being synchronized, rating tools designed to measure the technical skill and/or non-technical skill during the medical procedure may not be able to gather useful data on the mechanisms leading to adverse events/outcomes and establish correlation between performance and clinical outcomes.
According to some embodiments, measurements taken (e.g., error rates, number of adverse events, individual/team/technology performance parameters) may be collected in a cohesive manner. According to some embodiments, data analysis may establish correlations between all registered parameters if/as appropriate. With these correlations, hazard zones may be pinpointed, high-stakes assessment programs may be developed and/or educational interventions may be designed.
In an aspect, embodiments described herein may provide a device, system, method and/or computer readable medium for recording data which comprises multiple audio/video/metadata feeds captured by hardware devices in the OR/patient intervention areas (e.g., room cameras, microphones, procedural video, patient physiology data, software data from devices used for patient care, metadata captured by environmental/acoustic/electrical/flow-/angle/positional/displacement sensors and other parameters outlined herein). The captured data feeds may be simultaneously processed with an encoder (e.g., encoder 22 of
The data may be analyzed using encoder 22 (which may include analysis software and database) which preserves the time synchronization of data captured using multiple assessment tools/data parameters and allows export of the analyzed data into different statistical software. The exported data may be a session container file.
A device, system, method and/or computer readable medium according to some embodiments may record video, audio and digital data feeds from a clinical area in a synchronized fashion. The platform may be a modular system and is not limited in terms of the example data feeds described. Other data feeds relating to medical procedures may also be collected and processed by platform 10. For example, any measurable parameter in the OR (e.g., data captured by various environmental acoustic, electrical, flow, angle/positional/displacement and other sensors, wearable technology video/data stream, etc.) may be added to the data recorder (e.g., encoder 22 of
A device, system, method and/or computer readable medium according to some embodiments analyzes comprehensive, synchronized data using validated rating tools that consider different aspects or measurements of surgery/clinical interventions. These aspects or measurements may include: technical surgical performance, non-technical “team” performance, human factors, patient safety, occupational safety, workflow, audio/visual distractions, etc. Video, audio and/or metadata may be analyzed using manual and/or automatic data analysis techniques, which may detect specific “events” which may be tagged and time-stamped in the session container file or processed data stream.
A device, system, method and/or computer readable medium according to some embodiments records all tagged events on a master timeline that represents the entire duration of the procedure/clinical interaction. Statistical models may be used to identify and analyze patterns in the tagged events. The master timeline may be correlated to the processed medical data and the session file.
A device, system, method and/or computer readable medium according to some embodiments generates structured performance reports based on the captured and processed medical data for identification and determination of individual/team/technology performance measurements and organizational deficiencies that may impact patient safety, efficiency and costs.
A device, system, method and/or computer readable medium according to some embodiments provides a base for the design of targeted educational interventions to address specific safety hazards. These may include individualized training curricula, simulation-based training scenarios, Virtual Reality simulation tasks and metrics, and educational software.
A device, system, method and/or computer readable medium according to some embodiments may provide for high-stakes assessment programs for performance assessment, certification and re-certification.
Embodiments described herein may integrate multiple, clinically relevant feeds (audio/video/metadata) for a medical procedure, and allows a comprehensive analysis of human and technology performance for the medical procedure, organizational processes and links them to safety efficiency and outcomes as events, to develop solutions which aim to improve safety and efficiency and reduce costs.
Embodiments described herein may enable successful identification, collection and synchronization of multiple video, audio and metadata feeds relevant to a medical procedure (e.g., to evaluate different metrics of the medical procedure) with ample processing power to render all the video and audio in a useable fashion.
Embodiments described herein may employ measurement tools, and enable and incorporates objective assessment of various aspects of human and technology performance and environmental factors, with a view to understanding chains of events which lead to adverse outcomes in medical procedures and other aspects of medicine.
Possible applications for some embodiments include one or more of the following: (i) Documentation of various aspects of patient care in clinical areas with a high-risk for adverse outcomes. Comprehensive data collection by the encoder according to some embodiments may enable and/or provide for a detailed reconstruction of any clinical encounter. (ii) Analysis of chains of events leading to adverse outcomes. The data collection and processing according to some embodiments provide an opportunity to retrospectively evaluate one or more mechanisms and/or root causes leading to adverse outcomes in medicine and surgery. (iii) The analysis according to some embodiments may generate knowledge of the incidence and background of human errors and may enable development of strategies to mitigate the consequences of such errors. (iv) Design of training interventions for surgical teams. According to some embodiments, all identified crisis scenarios may be stored in a database and associated with simulation interventions which aim to prepare clinical teams for common clinical challenges and mitigate the impact of errors on clinical outcomes. (v) Evaluation/Improvement/development of existing/new healthcare technology and new treatments. According to some embodiments, the comprehensive data set may be used to evaluate safety hazards associated with implementation of new healthcare technologies. Furthermore, it may enable evaluation of the impact of healthcare technologies on efficiency. (vi) Use for certification and accreditation purposes. According to some embodiments, the data may be used for assessment of human performance and development of pass/fail scores using standard setting methodologies.
Embodiments described herein may be for use in association with OR settings. Embodiments, however, are not so limited. Embodiments may also find application in medical settings more generally, in surgical settings, in intensive care units (“ICU”), in trauma units, in interventional suites, in endoscopy suites, in obstetrical suites, and in emergency room settings. Embodiments may be used in outpatient treatment facilities, dental centers and emergency medical services vehicles. Embodiments can be used in simulation/training centers for education of healthcare professionals.
Example applications are presented for the purpose of illustration and are not intended to be exhaustive or to limit embodiments to the precise form disclosed. Other advantages, features and/or characteristics of some embodiments, as well as methods of operation and/or functions of the related elements of the device, system, method, platform and/or computer readable medium, and/or the combination of steps, parts and/or economies of manufacture, may become more apparent upon consideration of the accompanying drawings. Certain features of the system, method, device and/or computer readable medium according to some embodiments, as to their organization, use, and/or method of operation, together with further objectives and/or advantages thereof, may be better understood from the accompanying drawings in which present example embodiments. The drawings are for the purpose of illustration and/or description only, and are not intended as a definition of the limits of the invention.
Naturally, alternate designs and/or embodiments may be possible (e.g., with substitution of one or more components, units, objects, features, steps, algorithms, etc. for others, with alternate configurations of components, units, objects, features, steps, algorithms, etc.).
Although some of the components, units, objects, features, steps, algorithms, relations and/or configurations according some embodiments may not be specifically referenced in association with one another, they may be used, and/or adapted for use, in association therewith. The herein mentioned, depicted and/or various components, units, objects, structures, configurations, features, steps, algorithms, relationships, utilities and the like may be, but are not necessarily, incorporated into and/or achieved by some embodiments. Any one or more of the herein mentioned components, units, objects, structures, configurations, features, steps, algorithms, relationships, utilities and the like may be implemented in and/or by some embodiments, on their own, and/or without reference, regard or likewise implementation of any of the other herein mentioned components, units, objects, structures, configurations, features, steps, algorithms, relationships, utilities and the like, in various permutations and combinations.
Other modifications and alterations may be used in the design, manufacture, and/or implementation of other embodiments according to the present invention without departing from the spirit and scope of the invention.
Multi-Channel Recording Device or ENCODER
The multi-channel recording device 40 may receive input feeds 42 from various data sources including, for example, biometric sensors, feeds from cameras in the OR, feeds from wearable devices, feeds related to patient physiology from data stores, monitoring devices and sensors, feeds for environment factors from various sensors (temperature, decibel level, room traffic), feeds for device performance parameters, and so on. The multi-channel recording device 40 may synchronize and record the feeds to generate output data 44 (e.g., for export as a session file). The output data may include, for example, measurement values to assess individual and team performance, identify errors and adverse events and link to outcomes, evaluate performance and safety of technology, and assess efficiency.
There has been a paucity of research on contributing factors and underlying mechanisms of error in surgery. The complex, dynamic, and/or data-dense environment of the OR may make it difficult to study root causes of error and/or patterns of events which may lead to adverse outcomes. A synchronized multi-channel recording device 40 according to some embodiments provides a comprehensive overview or data representation of the OR. Modeled after the aviation black-box, this multi-channel recording device 40 or “black-box encoder” may register multiple aspects of the intraoperative OR environment, including room and/or procedural video, audio, sensors, an anesthesia device, medical/surgical devices, implants, and hospital patient administrative systems (electronic patient records). The black-box recording device 40 may be installed in real-life ORs/patient intervention areas at hospitals, outpatient clinical facilities, emergency medical services vehicles, simulation/training centres, among other places.
The black-box recorder 40 may be for use in anesthesiology, general minimally invasive surgery (MIS) surgery, interventional radiology, neurosurgery, and clinical practice. The black-box recorder 40 may achieve synchronization, audio, video, data capture, data storage, data privacy, and analysis protocols, among other things.
According to some embodiments, a multi-channel data recording device 40 is provided for use in the clinical environment which simultaneously records multiple synchronized data feeds, including procedural views, room cameras, audio, environmental factors through multiple sensors, an anesthesia device, medical/surgical devices, implants, and hospital patient administrative systems (electronic patient records). A multi-perspective view of the operating theatre may allow for simultaneous analysis of technical and non-technical performance and identification of key events leading up to an adverse outcome. Implementation of the black-box platform according to embodiments in real-life ORs may reveal valuable insights into the interactions which occur within the OR/patient intervention area, as a tool to identify, analyze and/or prevent errors in the intraoperative environment.
The multi-channel “black-box” encoder 40 integrates and synchronizes audiovisual/digital data feeds and/or other quantitative, semi-quantitative, and qualitative data feeds from a live OR or other patient intervention areas onto a single interface.
Hardware Unit
The encoder connects to one or more data capture devices that may be grouped as a hardware unit 20 (
The hardware unit 20 may be located the OR or other patient intervention area. For example, several pieces of recording equipment may be installed in the OR/patient intervention area, e.g., as follows: wall-mounted wide-angle lens room cameras to allow visualization of the entire room, several cardioid microphones to capture details of all conversation/noise/alerts in a quality that allows analysis, a procedural video capture device (endoscopic camera, x-ray, MRI etc.), and a vital signs monitor device and sensors (environmental, acoustic, electrical, flow, angle/positional/displacement and other), medical/surgical devices, and implants. The hardware unit (e.g., grouping of data capture devices) interface with middleware hardware devices and an encoder to connect and synchronize device feeds. Integration of the platform 10 may be non-intrusive in the OR, with minimal equipment set-up. The anesthesia and laparoscopic feeds may be streamed in the OR, and the microphones and room cameras may be installed without altering the infrastructure of the room, for example.
Room Cameras
According to some embodiments, hardware units 20 may have cameras 30 (
Microphones
According to some embodiments, hardware units 20 may have audio capture devices 34 (
Procedural Camera View
According to some embodiments, hardware units 20 may have cameras 30 (
Anesthesia Device
According to some embodiments, hardware units 20 may have patient monitor devices 36 (
In some embodiments, there may be extraction of raw digital data from the anesthesia device directly for provision to encoder 22 which ingests it as metadata.
Additional Sensors
According to some embodiments, hardware units 20 may have sensors 30 (
Hardware Unit Integration into the Operating Room
According to some embodiments, hardware units 20 may have a signal processor coupling data capture devices.
For larger application environments and to maximize efficiency and deliver increased fail-over and redundancy capabilities, a switching router may be used (e.g., router 16 of
For simplicity only one encoder 22 is shown but system may include more encoders 22 to collect feeds from local or remote data capture devices (of hardware unit 20) and exchange data. The encoders 22 may be the same or different types of computing hardware devices. The encoder 22 has at least one processor, a data storage device (including volatile memory or non-volatile memory or other data storage elements or a combination thereof), and at least one communication interface. The encoder 22 components may be connected in various ways including directly coupled, indirectly coupled via a network, and distributed over a wide geographic area and connected via a network (which may be referred to as “cloud computing”).
For example, and without limitation, the encoder 22 may be a server, network appliance, embedded device, computer expansion unit, personal computer, laptop, mobile device, tablet, desktop, or any other computing device capable of being configured to carry out the methods described herein
As depicted, encoder 22 includes at least one processor 90, memory 92, at least one communication interface 94, and at least one network server 12. In some embodiments, encoder 22 is configured for use as a perception engine 2000, among other uses and/or configurations.
Each processor 90 may be, for example, a microprocessor or microcontroller, a digital signal processing (DSP) processor, an integrated circuit, a field programmable gate array (FPGA), a reconfigurable processor, a programmable read-only memory (PROM), or any combination thereof. The processor 90 may be configured as described herein to synchronize the collected data fees to generate a container session file. The processor 90 may also implement anonymization and encryption operations, as described herein.
Memory 92 may include a suitable combination computer memory that is located either internally or externally such as, for example, random-access memory (RAM), read-only memory (ROM), compact disc read-only memory (CDROM), electro-optical memory, magneto-optical memory, erasable programmable read-only memory (EPROM), and electrically-erasable programmable read-only memory (EEPROM), or Ferroelectric RAM (FRAM).
The communication interface 94 may include an I/O interface component to enable encoder 22 to interconnect with one or more input devices, such as a keyboard, mouse, camera, touch screen and a microphone, or with one or more output devices such as a display screen and a speaker. The communication interface 94 may include a network interface component to enable encoder 22 to communicate with other components, to exchange data with other components, to access and connect to network resources, to serve applications, and perform other computing applications by connecting to a network (or multiple networks) capable of carrying data including the Internet, Ethernet, plain old telephone service (POTS) line, public switch telephone network (PSTN), integrated services digital network (ISDN), digital subscriber line (DSL), coaxial cable, fiber optics, satellite, mobile, wireless (e.g., Wi-Fi, WMAX), SS7 signaling network, fixed line, private network (including VPN 24), local area network, wide area network, and others, including any combination of these. These are examples of network infrastructure (e.g., network infrastructure 38 of
At 102, using the Customizable Control Interface 14 and GUI, a control command for activation of the system may commence recording, collection and streaming of all available audio, video and data feeds from data capture devices to one of multiple available encoders 22 via the switch router 16. The data capture devices may include a portion or all available cameras including both mounted and laparoscopic, all audio microphones and all available and implemented sensors and third-party devices (open or proprietary) used in a surgical unit, ICU, emergency or other clinical intervention unit. Pause/Stop/Play are additional control commands received at Control Interface 14 which may trigger transmission of corresponding commands to the encoder 22 to control recording.
At 104, in response to the control commands, data capture devices of hardware unit 20 capture data representing various aspects of the OR or other medical unit and generate feeds or data streams for provision to encoder 22. Various example data capture devices are described herein.
At 106, digital data may be formatted, translated and synchronized through middleware hardware and software and using networking protocols for clock synchronization across the network. Digital data will be ingested into the encoder 22 as Metadata.
At 108, the encoder 22 may be responsible for synchronizing all feeds to generate session recording, as described herein.
At 110, the encoder 22 may encode synchronized feeds into a signal transport file using lossless audio/video/data compression software. According to some embodiments, the encoder 22 may also be responsible for hosting (or storing) and operating anonymization and voice/vocabulary distortion software(s) for the purpose of protecting the identity of all medical professionals, patients, distinguishing objects or features in a medical, clinical or emergency environment. This may be done by encoder 22 either before compression, containerizing and encryption, or after decrypting in back office system.
Upon completion of the recording, at 110, the container file may be securely encrypted by encoder 22. Encrypt/decrypt keys may either be embedded in the master session container file and accessible through a master key, or have a separate key.
The encrypted file may either be stored on the encoder 22 (e.g., network server 16 of
According to some embodiments, the synchronized compressed encoded signals may be fed into a touchscreen monitor located inside the OR, which may be responsible for real-time visual display of feeds and direct recording onto an external hard-drive.
Control Interface
According to an embodiment, a user interface may be provided on a PC-based touchscreen monitor. The user interface may be referred herein as a Control Interface 14 (
According to an embodiment, the features of the Control Interface 14 providing the central control station (e.g., computer, tablet, PDA, hybrid, convertible) may be located in the clinical unit or another customer designated location. It contains a customizable graphical user interface (GUI) that provides a simple, user friendly and functional control of the system.
According to an embodiment, the Control Interface 14 may have a Play/Pause button. Some segments of the procedure may not need to be recorded. To skip these segments from the recording, the user interface may pause and restart the recordings when desired by way of control commands generated in response to activation of the play/pause button. The pause and play time-stamps may be recorded in a log the file, indicating the exact times of the procedure that were extracted.
According to an embodiment, the Control Interface 14 may have a Stop session button. When the “stop session” button is selected, files may be closed and automatically transferred to the storage area network (SAN), encoder 22, and so on.
According to an embodiment, the Control Interface 14 may have split-screen quadrant display of video feeds. Visual displays of videos may be provided in real-time during recording.
According to an embodiment, the Control Interface 14 may have a visual indicator of recording. For example; a red, blinking dot may appear on screen to provide visual indication to the team that video and audio feeds are being recorded.
According to an embodiment, the Control Interface 14 may have a log file. At the end of the recording, a log file may be generated that indicates key time points, including start and end of the recording session, pauses and replays.
According to an embodiment, the Control Interface 14 may have password protection. The interface may be secured with several layers of password protection to ensure maintenance of patient confidentiality and privacy.
Rich Content Analysis Unit (i.e. Video Analysis Software)
The Rich Content Analysis unit facilitates the ability to process, manage, review, analyze and tag multiple formats of rich content (for example, video, audio, real-time patient meta-data such as heart rate, and so on) in synchronization.
The Rich Content Analysis unit may provide, for the user (i.e. the medical professional, surgical expert or medical researcher), an intelligent dashboard which allows for the annotation and tagging of the rich content streams. That is intelligent dashboard may be an interview with playback viewing for reviewing content and interface controls for tagging content. The intelligent dashboard may be multi-dimensional in that the union of all dimension variables (i.e. case variables) may indicate a specific set of one or more applicable annotation dictionaries (i.e. coding templates). Some examples of the variables that may be used to determine the annotation and tagging dictionary may be: the type of medical procedure being performed (e.g., Laparoscopic Bypass), the aspect of the procedure that is being analyzed (e.g., technical skills, non-technical skills, and so on), the geographic area/region where the procedure is being performed (this may dictate a regional specific annotation dictionary that is mapped to a generalized globally accepted dictionary), and so on. These are example variables.
The Rich Content Analysis unit may implement a data model and cross reference between annotation dictionaries (i.e. coding templates) that span various medical procedures, country/regional interpretations, and so on. Each annotation dictionary may allow the entire rich content stream to be tagged (i.e. allows for the creation of descriptive content) in synchronization. For example, the content streams may be tagged with well-formed descriptors that are applicable to different objectives of analysis. For example, an annotation dictionary may allow for the tagging of Technical Skills (an example objective of the analysis) such as Suturing Error or Stapling Error (i.e. the tags) and tag every instance in the rich content stream where these types of errors may have occurred.
Rich content refers to multiple streams of content in various formats (audio, video, numeric data, etc.). The union of all Case Variables may require multiple annotation dictionaries—either custom made or based on previously validated rating tools—to assess different aspects of the procedure and recoding, including, but not limited too technical performance, non-technical performance, non-procedural errors and events, and human factors. Each annotation dictionary may be a well-formed relational dataset.
Another feature of the Rich Content Analysis unit is that the final aggregation of the entire rich content stream and the entire descriptive content (for example, the Technical Skills annotation/tagging, the Non-Technical skills annotation/tagging, and so on) can be reviewed in synchronization post aggregation.
The Rich Content Analysis unit may be disseminated with web technologies to ensure that the content is centrally hosted in a secure, healthcare institution approved environment. For each aspect of the procedure that is being analyzed, the Rich Content Analysis unit may ensure that only the applicable rich content streams are played simultaneously on a single user interface (for example, when rating the purely technical skills of the surgeon, the audio feed from the operating room would not be applicable). The Rich Content Analysis unit may provide numerous customizations that are again only made available depending on the aspect of the procedure being analyzed. These customizations include, but are not limited to: the ability to increase the granularity of any content stream (for example, enlarge or reduce the size of a video stream), control the playback speed of any content stream (e.g., increase or decrease the playback speed of a video), refine the quality of a content stream (e.g., apply filtration functions to increase the clarity of an audio stream).
Black Box Encoder Analytics Unit (i.e. the Black Box Database)
The Black Box Encoder Analytics unit may provide the second part in a two part handshake between the Rich Content Analysis unit. The Black Box Encoder Analytics unit may contain quantitative and qualitative analysis processes to facilitate reporting capabilities, including but not limited to, comparative analysis, benchmarking, negative trends, data mining, statistical reporting, failure analysis and key-performance indicators. The Black Box Encoder Analytics unit may also facilitate aspect based integration to statistical software research tools such as Matlab.
An example feature of the Black Box Encoder Analytics unit may be its relational database that captures and cross-references the entire dataset composition which includes, but is not limited to: the complete resultant annotated and tag content streams produced by the Rich Content Analysis software identified with structured meta-data such as the Technical Procedural Rating System for Laparoscopic Bypass, and so on; facility variables such as Department, Operating Room, and so on; procedure case variables such as urgency of the case, number of medical staff present and what their designation is, and so on; procedure case notes (in a structured well-formed relational data model) such as what kind of stapler was used, was hemostatic agent used, and so on; patient centric data such as blood work; and OSATS scores.
In addition to the example reporting capabilities listed, the Black Box Encoder Analytics unit may provide visual comparative analysis. The dataset can, in its entirety or a subset of, be displayed on a visual timeline that is distributed by relevant meta-data such as components of the annotation dictionary (e.g., Technical Errors) or Case Variables. Visual comparative analysis may provide example benefits, including but not limited to: the ability to review errors and events and determine preceding and trailing actions and observations; the ability to define, execute and convert visual observations into programmatic algorithms that can be executed on large groups of annotated content. For example, identifying, programmatically where a cluster of technical errors leads to a more serious technical event; the ability to baseline, benchmark, and refine inter-rater (i.e. content stream analyzer/reviewer) reliability by comparing timelines of different observers; the ability for medical teams to assess the cause of a major adverse event in a specific case—e.g., human error, medical device malfunction, and so on.
Another example feature of the Black Box Encoder Analytics unit is its dual purpose ability to improve patient outcomes with continuous improvement using healthcare intelligence analytics defined in the Black Box Analytics software. For example, the identification of small, unnoticed, possibly minor actions which may have led to a serious outcome; and support continuous improvement through additional research initiatives by integrating with research related software tools such as Matlab™ and providing research driven comparative analysis—for example, comparing a specific outcome using “Year 1” vs. “Year 2” research model.
Illustrative Example Applications
An illustrative example embodiment of the black-box recording device may involve: two wall-mounted high-definition wide-angled cameras; two omnidirectional microphones; a laparoscopic camera view; and a vital signs display. These are example data capture devices of a hardware unit. This example application may use an Internet Protocol (“IP”) network in which each data signal may be fed into an Ethernet switch (“ES”). The purpose of the ES may be to create a local area network (LAN) that establishes a central connection point for all sources. Before connecting to the ES, each data feed may be assigned its own Internet Protocol (IP) address. The video cameras and corresponding microphones may be IP-based with built-in encoders, while the laparoscope and anesthesia feeds may first run through an additional encoder device that converts the analog or digital video signals into a real-time streaming protocol (RTSP) video stream. The data signals may be bundled at the ES and directed to a touchscreen user interface on a PC-based platform (Patient Observation System, “POS”). The POS may be responsible for decoding the data into a readable signal, and synchronizing data feeds.
In some IP networks, video and/or audio feeds may be streamed separately through the network, from endpoint to endpoint, which may create opportunities for network delays along the streaming path. Over time, delays between video and audio feeds may accumulate, and/or each feed may experience different network delays. Delays may be unknown and/or constantly changing over time, and/or it may be difficult to quantify and/or account for delay and/or results in an effect called “drifting”. Another example embodiment of the black-box platform may be provided without the same IP-networking functionality of the example discussed above. Another example embodiment may use a self-clocking signal processor with synchronized micro-encoders. According to the example embodiment, the self-clocking signal processor may ensure that the audio and video streams are “locked” without drifting, and thus allowed the feeds to be shifted post-recording to achieve synchronization.
A further example embodiment of the black-box system may use omni-directional microphones, placed above the operating table and at the equipment boom, in an attempt to capture audio surrounding the surgical field. However, omni-directional microphones may have equal output/input at all angles, and/or may detect sound from all directions. These microphones may have resulted in suboptimal and/or inferior audio quality, with excessive background noise and poor detection of team communication.
In another example embodiment of the black-box system, directional cardioid microphones may be used which are sensitive at the front and isolated from ambient sound. These microphones may be placed on the infield monitor, directed towards the surgical field, where communication exchange may be likely to occur among the surgical team. This set-up may result in superior audio quality with clear detection of voices and sounds.
According to embodiments described herein, a synchronized multi-channel video/audio/metadata recording platform may be for use in the intraoperative environment. Development and installation of the black-box platform may be an iterative process that may involve both minor and major changes to the system.
While other industries such as television broadcasting may have equipment to capture video and/or audio, according to some embodiments, the “black box” platform for medical use may be cost-effective, ensure privacy of the patient and healthcare professionals, compact for storage in the OR, adapted for non-intrusive installation with existing equipment in the OR, designed to meet infection control standards of hospitals, and so on. Furthermore, the platform may integrate multiple feeds from multiple sources with multiple formats onto a single system, and may ensure that recordings are encoded to a common format that is compatible for subsequent data analysis.
The black-box recording equipment may have included one or more of the following: audio capture and synchronization and digital data capture. Integration of all these data streams may provide complete reconstruction of the clinical encounter. Communication may be a component of non-technical and human factors performance analysis. For example, communication failure may be a contributing factor to adverse events in the OR. Furthermore, team interactions in the OR may rely on verbal communication, which may not be properly evaluated without adequate audio quality. For example, for standalone video files, components of non-technical performance, including teamwork, leadership and decision-making, may not have been evaluated without an audio component. Audio may have been difficult to capture in the OR due to the multiple sources of noise within the room. Primary noise sources in the OR may include the following: preparing for operation (prior to incision), moving trolleys and equipment, doors opening and slamming, moving and dropping metal tools, suction, anesthesia monitors, alarms from anesthetic and surgical equipment, and/or conversation among staff and/or on the intercom. Microphone systems may be designed to capture all audio in the OR, for example: omnidirectional microphones to capture ambient sound, super-cardioid microphones to capture immediate surroundings of anesthetists, cardioid microphones to pick up conversations of clinicians in the surrounding area, and wireless microphones worn by anesthetists to capture their voices. While such a microphone set-up may be able to capture multiple noise sources, its intrusive nature in the OR may introduce a Hawthorne effect. Furthermore, mixing multiple audio feeds can result in poor audio quality, and analyzing each feed separately may be time-consuming.
According to some example embodiments, the platform may include an audio system with minimal microphones which produces optimal audio quality. For analysis of non-technical skills and human factors performance, team communication may be an audio source of interest. Since communication may occur at the surgical field, around the operating table, two cardioid microphones may be mounted on the infield monitors and directed towards the surgical team. An additional microphone may be set-up at the nursing station and directed towards the scrub nurse and equipment cart. A testing and validation phase may help microphone set-up. The testing may recreate noises of a surgical procedure in a real-life OR in order to identify a set-up that may result in a desirable and/or optimal audio quality.
According to some example embodiments, the black-box recording device also may provide both audio-video and multi-feed synchronization for proper data analysis. Audio and video feeds may be synchronized, as even a delay of one-thirtieth of a second, for example, between the two signals may create a detectable echo. Delay lags may increase exponentially over time. Example embodiments of the black-box recording device may have latency of less than one-thirtieth of a second, resulting in synchronization for proper data analysis. Multi-feed synchronization may be provided for multi-perspective analysis of a surgical case. The black-box device may enable the analysis of an event in the OR from multiple perspectives, such as for example, room view, procedural camera view, vital signs and digital data from various sensors. Latency between video/audio/data feeds may decrease the value of multi-channel video recording. In example embodiments of the black-box recording device, the digital data may be formatted, translated and synchronized through middleware hardware and software and using networking protocols for dock synchronization across the network. Digital data may be ingested into the encoder as Metadata. The encoder may be responsible for synchronizing all feeds, encoding them into a signal transport file using lossless audio/video/data compression software
For the design of recording equipment, the recording device may have a user-friendly interface which meets privacy concerns. The recording system interface may have a visual display of recorded feeds, among other things, to afford participants an awareness of the content of the recordings, and when recordings were happening. Furthermore, in some example embodiments, the recording equipment may be designed to maximize confidentiality and privacy of both patient and staff participants. Room cameras may be positioned to keep a patient's identity out of the field of view. Microphones may be placed to only capture communication around the surgical field, rather than off-the-record casual communication in the periphery. Some embodiments of the system may have a pause-feature which allows recordings to be easily and seamlessly paused during parts of procedures that are not meant to be recorded (e.g., intubation or extubation phases). Multiple layers of password protection may ensure that the recording system can only be accessed by authorized individuals from the research team.
The black-box may be built on the basis of a modular design—the recording system may be modified, feeds (and associated data capture devices) may be removed or added, without altering the primary/overall functionality of the system. This approach to design may allow for the black-box recording device or encoder to incorporate other data feeds and/or adapt to different clinical settings (e.g., ER department, ICU, endoscopy suites, obstetrical suites, trauma rooms, surgical/medical wards, etc.). The system may be modular, and may be expanded to accommodate for modifications and larger applications. The system may be able to incorporate additional video, audio and/or time-series data feeds (e.g., heart rate monitor, force-torque sensor) in other examples depending on the nature of the medical procedure and the available data capture devices.
“Black-Box” Data Recording Device in the Operating Room
The OR is a high-risk work environment in which complications can occur. Root-cause analyses may reveal that most complications result from multiple events rather than a single cause. However, previous efforts to identify these root-causes may have been limited to retrospective analyses and/or self-reporting. Example embodiments of the platform may implement a multi-channel data recording system for analysis of audio-visual and patient-related data in real-life ORs.
The “black-box” data recording device or encoder which, according to one or more embodiments, may capture multiple synchronized feeds in the OR/patient intervention areas: e.g., room and procedural view, audio, patient physiology data from the anesthesia device, and digital data from various sensors or other data capture devices. These feeds may be displayed on a single interface (e.g., control interface 14) providing a comprehensive overview of the operation. Data may be analyzed for technical skills, error/event rates, and non-technical skills. Post-procedure human factors questionnaires may, according to some embodiments, be completed by the operating team.
The black-box recording device or encoder may provide for analysis of technical and non-technical individual and team performance, errors, event patterns, risks and performance of medical/surgical devices in the OR/patient intervention areas. The black-box recording device or encoder may open opportunities for further studies to identify root-causes of adverse outcomes, and to develop specific training curricula to improve clinical organizational processes, and surgical/device performance, efficiency and safety.
Cloud Platform
Embodiments of the black-box recording device may address technical considerations improving synchronization, reducing latency exposure, providing extended and multi-zone modality and reducing over platform cost. A cloud platform may include the development of intelligent devices and generated time-stamps for the collected data for synchronization of devices and data.
The smart enabled devices and adaptors may autonomous or semi-autonomous intelligent devices including but not limited to smart cameras 1604, microphones 1606, data and media converters 1612, encoders 1610, adaptors and sensors 1608. In this illustrative embodiment, the smart enabled device or adaptor may incorporate and utilize a SOC device (system-on-chip) or FPGA device (Field Programmable Gate Array) in conjunction with on-board storage, power management and wireless radio(s). It may manage device requirements, device-to-device authentication, storage, communications, content processing, clock synchronization, and time stamping. Depending on factors, the technology may be integrated directly into the device or as an attached adaptor. In some example embodiments, the smart enabled devices and adaptors may connect directly to the CCS 1614 to provide data from the operating site via secure client network infrastructure 1618 and may receive data, commands, and configuration controls from CCS 1624 directly or via MMH 1612.
The black box encoder 1610 may be a composed of one or more computing devices, tablets and/or laptops which may run a secure user interface for the surgical staff to operate the black box platform. It may be resident on the client network connected via Ethernet or wireless (e.g., via station 1616) and may comply with the network security and IT policies. In some example embodiments, the black box encoder 1610 may connect directly to the CCS 1614 to provide data from the operating site via secure client network infrastructure 1618 and may receive data, commands, and configuration controls from CCS 1624 directly or via MMH 1612.
The Media Management Hub (MMH) 1612 may be a computing machine or server responsible for running the client media management software and its associated services. As an illustrative example it may run on Unix, Linux or Windows Server. The Media Management hub may be resident on the clients network and in addition to the necessary compute, IO and storage requirements, must be compliant to the client network security and IT policies.
Client Media Management Software (CMMS) 1620 may be an application running on the Media Management Hub 1612 that acts as an intermediate conduit between the back office central server and the smart enabled capture devices and adaptors. It may be responsible for the management and control of the black box platform resident on the client network. The CMMS 1620 may aggregate, package, compress and encrypt captured audio, video, medical device data, sensor data, logs, and so on. The CMMS 1620 may organize output files and categorizing by event using standardized file-naming conventions, keywords, file folders, and so on. The CMMS 1620 may provide device management including passing commands from the console, device authentication, security, file transfer hand-shakes, and so on. The CMMS 1620 has a device status dashboard with log file management and error reporting. The CMMS 1620 provides workflow automation, file management and transfer between the client site and the central server. The CMMS 1620 provides additional computing solutions with adherence to the client network security and policies. The CMMS 1620 provides processing and data transformation for clock broadcast for device synchronization.
Central Content Server and management software (CCS) Server 1614 may be located at a main site and act as two-way interface communicating with satellite or client site hubs. The CCS Server 1614 supports remote management, automation and file transfer hand-shakes for the delivery of packaged, compressed and encrypted content from client sites. The CCS Server 1614 acts as conduit to black box analytics software and databases as described herein.
High Performance Wireless Communications (HPWC) may be provided by one or more wireless stations 1616. For example, HPWC may be implemented using multi-gigabit speed wireless communications technology leveraging 802.11ad WGig, HD wireless, or prevailing standards in support of high-bandwidth digital content transmission.
A workflow is provided as an illustrative example of functionality. Upon receiving a command from a platform console located in the operating or surgical suite, the smart enabled device(s) will commence capture of the appropriate content (audio, video, digital data) to provide digital representations of the operating or surgical suite and people and objects therein. Smart devices or smart adaptors will process (e.g., record, store, generate, manipulate, transform, convert, and reproduce) the captured media and data, and embed a timestamp marker at precise timeline intervals in the output file.
The output files are transferred from the smart enabled device(s) to the MMH 1612 via Ethernet or High Performance Wireless Communication routers and/or devices, shown as wireless station 1616. Wireless routers may be multi-band wireless stations using 802.11ad or the prevailing multi-gigabit speed standards.
The CMMS 1620 may aggregate all media and data (audio, video, device data, sensor data, logs, and so on) and package, compress and encrypt to generate output files. Output files will be organized on network accessible storage devices using standardized file-naming conventions, keywords, file folders, and so on.
At scheduled intervals, files may be transferred over VPN tunnel (e.g., secure network infrastructure shown as client network 1618) from the client site to the processing facility or back office. The CCS 1614 at the receiving facility will manage file transfer and the distribution of content files, media and data to the black box analytics database.
The system 1600 implements synchronization techniques. For example, hardware-based encoding and synchronization may be implemented in part using software methodology. Data synchronization is conducted on the smart enabled device through the embedding of time stamps from the device clock. Device clocks are synchronized across the network via broadcast from the MMH 1612 over high speed wireless network (shown as client network 1618, wireless stations 1616, and so on). As synchronization is done at source by software, media and data may have near-zero levels of latency and the highest level of accuracy
The system 1600 implements device management techniques. Devices and coverage zones may be managed under administrative privilege on central console or remotely via the CCS 1614. Controls may be in place to prevent device scheduling conflict. The user may be presented optional capture configurations based on location, zone requirements or procedural type.
The system 1600 implements zone management techniques. As current hardware-based encoding and synchronization solutions are limited by the number of 10 ports available on the encoding device. Software synchronization and smart enabled devices may allow for greater scale and ease of deployment. Extended zone and multi-zone captures can be attained thereby allowing for richer content and longer visibility to chain-of-events in support of the data analysis.
The system 1600 implements device status techniques. For example, smart enabled device or adaptor operating status will be broadcast from authenticated devices back to the CMMS 1620. Administrators at client site and/or remotely through the CCS 1614 may be able to access a device dashboard interface that automatically generates visual representations of data reporting key operating metrics and statuses on all authenticated smart enabled devices (e.g., on-line, off-line, running capture, on-board storage, and so on). Where a smart enabled device or adaptor is operating outside of normal conditions (e.g., storage full, off-line) then an alert (email, SMS) will be transmitted to the administrator and appropriately logged.
The system 1600 implements file management techniques. Upon completion of capture and processing on the smart enabled device or adaptor, processed files will be transferred to the MMH 1612. The CMMS 1614 will communicate with the device and transfer will be confirmed via hand-shake. Each device or adaptor may have on-board storage which will serve as short-term file redundancy and recovery across the platform.
The system 1600 may provide reduced cost, lower latency, and higher flexibility. Multi-core encoders and copper cabling in restricted workspace may translate to high costs and commissioning complexity. Cable routing has to be pulled through conduit in sterile core. Cable lengths impact latency of signal. Hardwired connections may restrict device placement and impact capture quality. Example embodiments described herein may be based on a software solution (at least in part to configure various hardware components), over wireless, and using smart enabled devices may reduce overall hardware cost, yield higher accuracy and capture quality, greater flexibility, and ease of commissioning.
Motion Tracking
Embodiments described herein may implement motion tracking using 3D cameras or IR devices. For example, the black box platform may collect and ingest motion tracking data for people and objects at the surgical site. To maintain complete freedom in a clinical environment, markerless motion tracking may be required. Data may be collected from 3D cameras or time-of-flight cameras/sensors.
The platform may implement motion tracking techniques using various components and data transformations. For example, the platform may include one or more autonomous or semi-autonomous 3D depth cameras or Time-of-Flight (TOF) sensors using laser and/or infra-red (IR) devices. As another example, the platform may generate distance and/or position information from the output signal of the TOF sensor and that it converts into a 3D depth map or point cloud. Embodiments described herein may include a computing device for processing output data from 3D camera or TOF sensor. Embodiments described herein may provide customized data processes to distinguish motion resulting from changes in captured depth maps. Embodiments described herein may provide media management hardware and software to aggregate, package, compress, encrypt and synchronize captured point clouds as motion data with other collected media. Embodiments described herein may provide a Central Console for device and capture management and processing software to convert motion data into analyzable information to be used in study of human factors, workflow design and analysis of chain-of-events.
A workflow is described to provide an illustrative example of functionality provided by the platform. In some examples, 3D depth cameras or TOF sensors are fix-mounted in the operating or surgical suite. On receiving a command from the platform, the cameras capture and generate distance and position information of the viewable capture area. Output data will be passed to a computing device running a custom process that creates and establishes a baseline measurement (static field map) and provides summarized motion data by comparing and measuring changes in position information between adjacent 3D depth maps and point clouds. The collective baseline and frame measurement data may be passed to the Media Management Software (e.g., software 1620 on MMH 1612) which may aggregate, package, compress, encrypt and synchronize motion data with the other collected media.
At scheduled intervals, files will be transferred over VPN tunnel from the client site to the processing facility or back office where the motion data will be processed into analyzable information to be used in study of human factors, workflow design and analysis of chain-of-events.
An example process may involve different operations, including for example, a compute operation to receive 3D depth maps or point clouds formatted and structured to be able to conduct point-to-point measurements of change. The compute operation may then create and establish a baseline measurement (static field map), and analyze and record changes in adjacent depth maps or point clouds. The compute operation may map changes to a common timeline and summarize change data on a time continuum basis for purposes of comparison to the reference static field map.
Embodiments described herein may provide synchronization of devices and collected data. For example, the platform may implement synchronization of various media streams to a common timeline as a factor in the determination of the quality of analytics. The following is an example of requirements to maintain accuracy in synchronization: direct connection between all source devices into a general purpose computer; sufficient 10 and compute power to compress, encrypt, encode and organize multiple streams of audio, video and data files; an assessment, determination and understanding of latency for all incoming feeds; utilities or algorithms to tune and calibrate infeed's of data to insure synchronization (example introduce offsets); and calibration of time stamps in file headers to a common standard for playback.
Embodiments described herein may provide analytics tools. In future embodiments, process operations may translate point cloud and/or depth mapping position, distance and change measurements into real-world distance measurements. These measurements may permit the creation of the key performance indicators (KPI's), in a semi-autonomous fashion. KPI's can be used to further analysis and/or provide recommendations on workflow and human factors impacting timeline and chain of events. These may include: steps taken, distance travelled, pathway taken vs optimal pathway, impacts of unintended collisions or clustering, impacts of spatial design, impact of arrangements and orientation of staffing, equipment, devices, and so on.
Analytics Applied to the Black Box Data Set
Embodiments described herein may implement data-driven surgical error analysis tools to investigate mechanisms of errors, and to assess error and event patterns. Embodiments described herein may implement process operations for formative feedback, self-assessment, learning and quality control, and to identify patterns, correlations, dependencies and signatures from data collected.
Embodiments described herein may provide an application of data-driven modeling to identify, and extract features, correlations and signatures from data collected and analyzed from the OR black box encoder. Data-driven modeling offers a sound perspective to describe and analyze all those systems for which closed-form analytical expressions may be difficult to determine. Using datasets of input-output pairs of samples related to the problem, the objective is to use Computational Intelligence (CI) to reconstruct a mathematical model that recognizes key factors and predicts clinical outcomes, costs and safety hazards. CI tools may include neural networks, support vector machines, fuzzy inference systems, and several techniques from time-series analysis and dynamical complex systems. Using Cl-based approaches, both offline and online solutions could be built for analyzing errors, adverse events and adverse outcomes in surgery. The term offline refers to solutions that may be used to automatically infer knowledge (e.g., rules of causations, correlations) from examples describing past events recorded in the OR. The online approach may provide a real-time tool to assist surgeons and OR teams intra-operatively. Such an instrument may operate by monitoring the current conditions in the OR, reporting events that may lead to conditions of potential errors (e.g., the noise level, temperature, number of individuals in the room, and so on).
The following provides an overview of computational intelligence methodologies applied in the OR black box encoder solution. Computational intelligence methodologies may be used to design networks capable of extracting features, correlation and the behavior of events that involve complex, multi-variable processes with time-variant parameters. For the present application, methods may include artificial neural networks (ANN), both feed forward and recurrent, radial basis function networks (RBFN), fuzzy logic systems (FLS), and support vector machines (SVM). Applied to the data generated by the OR black box, these systems will be capable of implementing various functionality, including for example, finding complex, nonlinear and hidden relationships among the data representing human performance, patient physiology, sensors, clinical outcomes and clinical costs, and predicting outcomes and behaviors. Further example functionality includes a functional generalization and, as such, acceptably responding to situations to which the OR black box encoder solution has not been exposed before, and offering alternative solutions when the system cannot be expressed in terms of equations, or when a mathematical model does not exist or is ill-defined.
Example advantages of FLSs are the capability to express nonlinear input/output relationships by a set of qualitative if-then rules, and to handle both numerical data and linguistic knowledge, especially the latter, which may be difficult to quantify by means of traditional mathematics. The main advantage of ANNs, RBFNs and SVM, on the other hand, is the inherent learning capability, which enables the networks to adaptively improve their performance. The present solution may apply CI methodologies, including ANN, RBFN and SVM, to develop robust networks and models that will extract features, detect correlations, and identify patterns of events from the OR black box dataset.
As noted, embodiments described herein may implement data analytic techniques using artificial neural networks. For example, time-series modeling may include applications of time delayed ANNs and feedforward multi-layer perceptron networks to model nonlinear dynamical systems. As another example, hybrid stochastic and feedforward neural networks may be used to predict nonlinear and non-stationary time series by incorporating a priori knowledge from stochastic modeling into neural network-based predictor. As a further example, two-layer neural networks consisting of a series of nonlinear predictor units together with a Bayesian based decision unit for time series classification. As another example, ANNs for time-series prediction and the impact of the use of the heuristics to select the optimum size of the sampling window. Other neural network topology may be used, such as a recurrent architecture whereby temporal relations can be built into the network via feedback connections. Recurrent neural networks have been extensively investigated for periodic and chaotic time-series prediction. A few additional examples include applications of robust learning operations for recurrent neural networks based on filtering outliers from input/output space suitable for time series prediction; various selection methodologies for optimal parameter adjustment in pipelined recurrent neural networks used for prediction of nonlinear signals; complex-valued pipelined recurrent neural networks for modeling/prediction of nonlinear and non-stationary signals; recurrent predictor neural networks in combination with self-adaptive back-propagation through time learning algorithm for prediction of chaotic time series; and self-organizing map and recurrent neural networks to model non-stationary, nonlinear and noisy time series.
Some example embodiments may use radial basis function networks where feedforward and recurrent RBFNs may be examined for time-series modeling of the black box data sets.
Some example embodiments may use neuro-fuzzy networks. Different adaptive neuro-fuzzy inference system (ANFIS), alternate neuro-fuzzy architecture (ANFA), dynamic evolving neural-fuzzy inference system (DENFIS) to chaotic time series prediction may be utilized. Examples of such application include: (1) real-time neuro-fuzzy based predictors for dynamical system forecasting; and (2) hybrid recurrent neuro fuzzy networks using non-orthogonal based wavelet, recurrent compensatory neuro-fuzzy systems, and weighted recurrent neuro-fuzzy networks for modeling of nonlinear dynamic systems.
Further example embodiments may use support vector machines. The SVMs may be used for time-series forecasting of clinically-relevant performance outcomes, adverse events, complications and costs/return on investment.
Some example embodiments may use nonlinear Black Box data modeling techniques. In cases of an absence of a priori information, embodiments described herein may use a model that describes the dynamic behavior (features/signatures) of the system on the basis of a finite set of measured input-output pairs. Various nonlinear black-box modeling problems can be realized as that of selecting the best mapping mechanism using the input-output data and then trying to minimize the error between the output of the model and the measured output.
In some embodiments, smart data analytics techniques may be applied to transform unstructured data into structured and meaningful information. The framework to analyze the black box data may be focused on data-driven analytics and data modeling. The OR black box may be provided to develop medical domain expertise that may be used to customize the data analytics tools to the specific healthcare needs.
A framework is established to assess performance and medical errors and adverse events that are observed in the black box data (e.g., data obtained by encoder 22). This assessment framework has been developed by leading global healthcare practitioners and follows leading practices in the field.
The technical performance of the surgeon is assessed by using an error-rating tool and a global rating scale. In addition, the intraoperative team performance of the surgical, anesthesia and nursing team are assessed using standardized (e.g., normalized) rating tools.
A step includes the provisioning of a robust protocol for data extraction into a relational database management system (RDBMS) with a well-defined entity-relationship diagram. The black box data (e.g., data provided to encoder 22 in the form of various data stream feeds) may include video data (in-room videos and the procedural video), audio data (directional microphones and decibel level in room), physiological patient data, signal output data from surgical instruments, room temperature data, and foot traffic data from motion sensing input devices, etc., among others. These feeds may be provided at different times and may require synchronization by encoder 22 prior to extraction. In some embodiments, the feeds already have features extracted and provided in the form of machine-readable and/or interpretable formats, in other embodiments, the feeds may first require processing or pre-processing to extract feature sets.
The volume, velocity and variety may be similar to “big data” and “big data” techniques may be applicable. The metadata may be stored in various formats, such as in records stored in MySQL. In some embodiments, more advanced storage systems such as the Hadoop Distributed File System (HDFS) may be used to store the video data.
To extract the data in a meaningful and structured way, time stamping may be used to provide temporal order among a set of events. Time-stamped events may be displayed on a timeline, and summarized in a matrix that indicates the duration and frequency of each event. The timeline provides a visual representation of time-stamped events to facilitate data analysis and the development of predictive algorithms.
Another step may include the exploratory analysis of the data. There may be various features collected in the data, and exploratory analysis may include utilizing data statistical analysis to identify and extract features. The temporal nature of the data may lead to the data being particularly well suited for time series analysis to extract meaningful statistics and insights. Time chart, distribution, autocorrelation, cross-correlation and spectral analysis may be used to investigate the temporal nature of the data.
The exploratory analysis of the data holds the opportunity for pioneering work in the field of data analytics of Black box data. By combining data analytics with domain expertise, questions such as “What are root causes of negative outcomes?” may be answered.
Root cause analysis (RCA) is a framework that may provide a proven method of problem solving used to identify the root causes of errors. Hierarchical clustering models in data mining may be used for root cause analysis to determine cluster groups and events. The output of the hierarchical clustering models may be a dendrogram. The 5 ‘whys’ may be asked to identify the causes associated with each sequential step towards the outcome. Causes may be classified into two (or more) categories: causal factors and root causes. The root cause category has the defining characteristics that it interrupts the sequence chain when eliminated, and the outcome of the root cause analysis may provide invaluable insights to complement medical insights in understanding the cause of negative and positive outcomes.
Data-driven modeling may be used to describe and analyze systems, even when a closed-form analytical expression is difficult to determine. An objective of data modeling is recognition and prediction of the relevant patterns that cause negative outcomes. This is where cognitive computing and smart analytics may be utilized.
The data modeling toolbox leverages the machine learning capabilities. While the use of hierarchical clustering for root cause analysis (RCA) is previously discussed, there may be other supervised learning techniques that some embodiments may utilize, such as support vector machines (SVM) as well as artificial neural networks (ANN).
Support vector machine (SVM) approaches offer a robust approach to classification originally developed as a linear classifiers and later expanded into a nonlinear classifier using the kernel trick. Artificial neural networks (ANN) may also offer a powerful approach for relevant applications that include time series prediction, clustering, classification and pattern recognition. The approaches may be used individually or in combination, according to various embodiments.
SVM and ANN excel at finding hidden relationships in data-rich content as well as providing predictive behaviour of the system. The ability to model, recognize and predict the relevant patterns has tremendous potential not only for historical data but also for real-time data analytics. In some embodiments, domain expertise, data modeling and machine learning techniques are applied to the black box data (e.g., stored in encoder 22).
This may lead to: identification of frequent temporal patterns leading to errors/adverse events in the timelines, and development of predictive algorithms that can identify critical events during surgical procedures.
The timeline data collected in parallel may, in some embodiments, contain time-stamped temporal events.
The time-stamped series of technical and non-technical events may be recorded and categorized depending on: if the events were caused by nurses or surgeons, or other factors. Data analytics tools on the timeline analyzed by the medical team such as N-gram pattern extraction algorithm may be used, in some embodiments.
In some embodiments, MinEPI pattern extraction algorithm(s) may be used for identification of parallel and hybrid patterns. Patterns of events and their relationships with errors may then be validated and incorrect patterns may be filtered out.
This temporal pattern recognition may be useful in automating the process of determining the events and errors commonly leading to adverse events/adverse outcomes.
To support the data extraction the exploratory analysis and to exploit the advanced modeling and machine learning capabilities of the black box system, a platform may be provided for aggregation and capture of relevant and specific content used in the systematic analysis of operating and medical procedures. The extensive volume of varied data accompanied by its veracity may pose specific challenges to scale and efficient processing.
Therefore, any platform used must be scalable, robust, and provide the necessary capabilities to intelligently filter, annotate and correlate events and behaviors for use in the extraction and the analytics phases. In some embodiments, the black box system leverages a medical centric perception engine 2000, which, as noted in
Various types of content and data captured by the black box system across multiple sources, inputs and device types including but not limited to: cameras (optical, 3D depth, laser, stereo), microphones, sensors (examples include but not limited to acoustic, environmental, flow, positional, displacement, electrical, and other sensors), autonomous and semi-autonomous smart devices, intelligent surgical instruments (e.g., smart staplers, intelligent laparoscopic instruments, etc.), surgical equipment (e.g., anesthesia monitors, patient monitors, energy devices, O2 saturation monitors, etc.), pre-operative monitoring systems, post-operative monitoring systems, patient record systems and databases, wearables. etc.
Interconnection of data sources may be provided through direct and indirect services, using wired, bluetooth, near field (NFC), Li-Fi, prevailing wireless protocols, etc. These services may act in a coordinated fashion as part of a device mesh and/or private client network.
Devices may use available on-board communication or for those devices without the necessary communication structure, through smart enabled adaptor modules which act as autonomous agents connecting the device to the network.
Smart enabled adaptors may provide network connectivity, security and processing capability (e.g., record, store, generate, manipulate, transform, convert and reproduce). Captured data may be time-stamped, geo-stamped, or otherwise classified, categorized or profiled at intervals. Security not only may also be designed to manage access and control, to block attacks and measure threats, but also to deliver advanced threat detection, response and counter-response.
A perception engine 2000 may be provided to address challenges faced in scale and data processing efficiency, adopting technologies and strategies from IOT (Internet of Things), 10E (Information of everything), adaptive technologies, advanced machine learning methodologies, and computational intelligence. Advanced algorithmic controls and software based decision-making may be deployed at targeted nodes and may act as a black box perception engine 2000. Nodes may be local, regional, or global and are otherwise characterized as locations where decisions and filters are made.
The perception engine 2000 may be configured to filter content, categorize, profile, extract features, uncover underlying data behaviors and provide evidence of correlation of events in complex multi-variable processes and timelines. In some embodiments, the perception engine 2000 is configured to prioritize and/or weight the various feeds and/or features for analysis, such that perception engine 2000 is tuned and/or biased towards feeds and/or features of particular importance. Such determination for biasing the perception engine 2000 may be based on statistical reviews of the accuracy (e.g., sensitivity, specificity) of the results of perception engine 2000. The perception engine 2000 is configured to generate predictions and/or determinations based on analyses of various extracted features, estimating when clinical events (e.g., technical errors, technical events) have occurred based on information obtained from the data sets. The specific approach taken by the perception engine 2000 is, in some embodiments, configured to modify itself over time (e.g., modifying rules, weightings, splitting criterion) and iterations with data sets such that perception engine 2000 becomes tuned in developing predictions. In some embodiments, a high (or baseline) specificity rate may be targeted as an outcome as the predictions iterate through incoming data sets.
In an embodiment of a two-node implementation, connected devices, sensors, record systems, equipment and machines, may be networked with a computing platform that operates software-based controls and algorithms utilizing some type of neuromorphic processor (e.g., IBM TrueNorth™ or similar neuromorphic device). The devices and computer may operate on a private network.
Incoming inputs from source may be fed to a computing platform where they are monitored. Onboard software in conjunction with a neural processor may be configured to conduct first level filtering, making criteria-based decisions, extract identifiable features, categorize events, make inferences on data sets, conduct classifications of data and data streams.
The software algorithms and actionable decision matrices used by a computing platform may be developed based on clinical science, advanced statistical analysis and evidentiary research, among others. The programming may be conducted in a language compatible with a neuromorphic processor, generating something similar to spikes, and relationships of input/output, cause/effect, decision making etc., may be mapped, weighted and transcribed in the form of a firmware or program. Some embodiments may operate as neurosynaptic core or array of cores like a TrueNorth™ neural processor and use their architecture of axons (inputs), synapses (connections) and neurons (computations/outputs).
Filtering of inputs and feeds from devices may be processed in real time. Classifiers and decision making may be conducted utilizing developed algorithmic libraries and computational intelligence. Accuracy and predictive capabilities may be improved through back propagating and machine learning techniques.
Network based software may manage filtered content including storage, routing, archival, local node and network security, and may be responsible to transmit data and filtered information to the next node over a secure private network. On receipt, filtered content may be added to the post-processing databases adding and increasing to the overall evidentiary knowledge base.
Additional content analytics, comparative analysis, hazard maps etc., may be conducted. Correlated linkages may be identified with this filtered content. Increased correlation, previously unidentified or underlying data behaviors across statistical datasets, and further refinements and predictors may be fed back into the perception engine 2000, thereby increasing the refinement of filtering algorithms moving forward.
Some examples and embodiments describe capabilities and logical flows of the platform, an ability to address scalability and processing efficiency and the exploitation of modelling and machine learning for advanced analysis. The embodiments may range from simple to complex multi-nodal applications.
Steps in a process for providing machine learning using perception engine 2000 may include training the algorithm through a manual review of feeds, marking points in time where clinical events have occurred, and the manually flagged information may be provided into the perception engine 2000 for training purposes. The perception engine 2000 may then utilize machine-learning approaches to re-process the data feeds to generate predictions for additional errors based on the extracted features and patterns, generating outputs indicative of additional errors linked to timestamps stored on a common timeline. Accordingly, in some embodiments, the data set required for training may need to be a sufficiently large data set.
The perception engine 2000 may be tuned such that clinical events are flagged with a particular confidence level and/or confidence score. The confidence level and/or the confidence score may also be associated with a competence level or a competence score. While the perception engine 2000 itself can be modified in terms of which features to focus on and/or to weigh on, the perception engine 2000 results may also be filtered based on confidence level and/or confidence score. In some embodiments, the confidence level and/or confidence score associated with various clinical events is represented graphically on an output, such as an annotated timeline chart. In some embodiments, the confidence level and/or confidence score is stored in metadata and incorporated into instruction sets for notifications of when in a particular surgical procedure the data feeds should be reviewed to assess the presence and/or absence of technical errors and/or events.
In an embodiment, output layers of a one perception engine would be combined with other inputs or outputs from other perception engines to form a multi-node perception engine. Such a configuration could provide even higher order processing and richer content to be used in advanced analytics, statistical processing and problem identification and hazard mapping (e.g., heat maps, confidence levels, potential for clustering of events).
Another embodiment includes a wide area perception engine with aggregation across spatial or logical dimensions. Across high speed networks mapped use cases may allow for extraction of features and data relationships across spatial or logical dimensions thereby allowing for comparisons between different aggregation points.
Such a system may extend the black box system beyond local assessment and provide insight and evidence across healthcare system levels, between regions and benchmark against global standards for the exploitation of leading practices on a macro level.
The OR black box system may be designed to focus on developing targeted interventions based on the obtained data to address performance and organizational deficiencies, and the most frequent hazards to patient safety at an individual, team and institutional level.
Currently, the vast majority of safety regulations and training interventions are based on retrospective root-cause analyses, or have been derived from other high-risk industries that may or may not be relevant in health-care.
Some embodiments of the system may be adapted to target the development of evidence-based interventions using prospective data that has been generated at the point of care, and this concept has never been used in healthcare before.
The data obtained from the systematic analysis of operative procedures may provide insight into the complex processes within the healthcare system, allow assessment of performance on an individual and team level, and evaluate human interactions with modern technology.
Furthermore, the data can be used to determine specific individual and team weaknesses, hazard zones (e.g., through heat maps) within procedures as well as characterize the cascade of events that result in “near misses” or adverse patient outcomes. The information may deliver knowledge content useful to tailor effective educational interventions based on real life observations rather than hypothetical scenarios used in current training. The concept, in relation to experiential learning may be used to create educational strategies that can be disseminated to sites that do not have access to their own real-life data.
The process may include the following steps:
In many cases the cause of adverse patient outcomes remain elusive since they are frequently multifactorial and based on retrospective analysis. By analyzing all prospectively documented adverse outcomes using the OR black box system, patterns of recurrent problems will be identified, characterized and used to generate a set of training scenarios based on the real experiences. This knowledge may relevant to healthcare personnel (OR teams) involved in patient treatment in similar clinical contexts.
The educational content can be compiled to information sheets, textbooks, e-learning software and integrated into standard operating procedures (SOP) at an institutional level. Beyond summarizing common or significant root-causes of adverse outcomes, these scenarios may be used to generate software packages for full-scale simulations in virtual operating rooms. The variables may be programmed into the simulation software and thus be packaged, commercialized and exported to educational institutions.
Surgical errors are “symptomatic” and can be caused by different factors, only one of these being skill level of the surgeon. Errors may be the result of lack in technical skill (as observable in surgeons in training), lack of judgment and cognitive factors (as observed in procedures by surgeons in training/inexperience), and the complexity of the case or surgical step (observable in expert procedures). Experimental results indicate that errors occur in clusters. In order to eliminate confounders, procedures performed by expert surgeons may be examined where the error clusters would be indicative of “complex” steps and where educational interventions should be developed to improve performance.
Through the analysis of the data generated by the OR black box system, error distribution patterns for numerous procedures and surgeon training levels may be identified. These specific patterns may be used when designing curricula focused on creating “error awareness” and in conveying mitigation concepts.
Mapping procedure complexity and identifying potential hazard zones may be used to create educational strategies targeted directly at these steps. Instructional strategies such as but not limited to deliberate practice may then be used to train for these steps and thus minimize the risk of adverse events. Informing surgeons about complex or hazardous steps also enables the design of standard operating procedures, which is common in aviation (for example the “sterile” cockpit concept during takeoff and landing), to limit distractions during these sensitive steps (no procedure irrelevant conversation, minimize room traffic, reduce overall noise level).
Recognition of specific behavior patterns within teams that are either beneficial or detrimental to patient outcome is a step that may be used to subsequently fashion specific team training interventions and debriefing sessions. The data generated through the OR black-box system 1600 observations may be used to identify specific patterns in non-technical performance of the teams. This information may serve as the basis for the design of specific team interventions using OR simulations, role-play and debriefing sessions. Recurrent themes that are identified as affecting team performance and processes on an organizational level may be addressed by policy recommendations and the design of standard operating procedures.
At an individual level, deficiencies in technical and non-technical performance may be addressed in targeted interventions utilizing evidence based instructional strategies and approaches such as but not limited to: behavior modeling, debriefing, warm-up training, deliberate practice and simulation.
At the level of the OR team, inter-professional teaching sessions and training interventions may be tailored to the most common observed team failure patterns and identified threats to patient safety. Evidence-based methods such as but not limited to crisis-simulations, in-situ simulations, role-play, and group debriefing may be employed.
At the organizational level, measures may include policy changes to address operation room ergonomics, OR processes, and environmental factors such as noise and OR traffic. In addition, the data may be used to design specific checklists and standard operating procedures for the most relevant or hazardous situations.
Machine Learning Example
In some embodiments, the black box platform may be utilized in relation to tracked data in the context of an operating room.
For example, relationships may be stored in the form of attribute-relation file format files (ARFF), among others. Experimental data was recorded from 31 cases/surgeries that occurred between May 2015 and August 2015, and audio recordings were analyzed. 559 technical errors (e.g., where a surgeon made an error such as applying too much pressure to a patient's intestine) were tracked, where there were 207 technical events (e.g., where the pressure on a patient's intestine caused the intestine to rupture). Two lengths of recording clips were made, a first set at 30 seconds, and a second set at 30 seconds. Voice recordings were made 5 seconds before error and 25 seconds after error for 30 second recording clips, and made 10 seconds before the error and 50 seconds after the error for 60 second sound recording clips.
Features tracked in the sound recordings included: volume mean, standard deviation, volume dynamic range, average magnitude difference function (AMDF) mean/standard deviation, short time energy mean/standard deviation, spec centroid: mean and variance, spec roll: mean and variance, spec flux: mean and variance, zero crossings, and the first 5 Mel-frequency cepstral coefficients (MFCCs), 1 2 3 4 5.
Audio was extracted from multiple cases at once, and features were often extracted overnight. A software package was used for classification and attribute selection. Through experimentation, Applicants found that decision tree ensemble methods provided better accuracy and a 10-fold cross-validation. Table 1 is a table of audio info results:
Features were removed, and experimental results indicated that for a removal of 7 features, a random forest approach yielded a rate of 63.059% to 64.9374% when 7 features were removed. An Alternating Decision (ADTree) approach yielded a rate of 58.05% to 58.67% when 8 features were removed.
As provided in the above results, Applicants noted that not all features were equally helpful in determining the presence of a clinical event. Accordingly, in some embodiments, the perception engine 2000 may be configured to remove various features from consideration, and in some further embodiments, the perception engine 2000 may be configured to automatically remove various features from consideration in view of an automated decision process wherein contributions are analyzed using statistical methods (e.g., p-test) and features below a particular contribution level are excluded. Such an approach may be advantageous especially where it is unclear as to which features are better or worse than others.
In an embodiment, tracked features include at least one of a volume standard deviation, a volume dynamic range, an AMDF mean, an AMDF standard deviation, a short time energy mean, a short time energy stand deviation, a spec centroid mean, a spec centroid variance, a spec roll variance, a spec flux mean, a spec flux variance, and a fifth mel-frequency cepstral coefficient. In some embodiments, the ADTree is not limited to audio features, but can also be utilized for biometric data features and video data features.
In another embodiment, tracked features include at least one of a volume standard deviation, an AMDF mean, a short time energy mean, a spec centroid variance, a spec roll variance, and a spec flux variance. These specific features were found during experimentation to be particularly relevant.
The analysis may be re-run on various data sets to determine whether such removal was indeed beneficial from a perspective of increasing sensitivity and/or specificity. Applicants, through experimentation, found that some embodiments were particularly accurate in identifying clusters of clinical events (e.g., clusters of technical errors).
An ADTree was found to have various benefits, such as being interpretable, easy to implement, providing an inherent way to ‘control’ sensitivity/specificity and being consistently one of the more accurate classifiers (e.g., it was found to be accurate in experimental iterations).
In using an ADTree, the system is configured to find various conditions throughout an entire tree that minimizes a splitting criterion, for example, analyzing DKM instead of entropy, 2*sqrt(p(1−p)) instead of SUM (−p log 2p), concavity, among others.
Scores may be found based on natural log of the ralo of +ve and −ve examples, for example, 0.5 In (positive/negative). Rules may be added to classifiers, and continued until an unspecified stopping criterion.
Prediction timelines were created with a ADTree algorithm and training data, and the process included extracting surgical audio, identifying features from specific cases' surgical audio, running features through an ADTree classifier, writing predicted times into a timeline, and copy and pasting a timeline into existing rated timeline to qualitatively compare results of prediction. In some embodiments, a ADTree threshold may be amended to modify and/or balance between sensitivity and/or specificity of the system. Table 2 is an example table illustrating some experimental results based on an embodiment.
As indicated in the results of
The results however, may be indicative that the embodiment of the system was able to estimate some clinical events as noted as “potential errors” in proximity to actual technical errors, technical events as determined by surgical raters. The results further indicate that the system was more accurate when technical errors occurred in clusters. The removal of features may be particularly important where some of the features originally thought to be useful may actually be misleading and detrimentally impact the training of the system. For example, some features may be helpful from the perspective of providing useful splitting criterion, while other features may simply provide “bad data” where patterns cannot be ascertained.
As noted in
Applying the outputs of the perception engine 2000, the system may be able to generate notifications and/or other control commands indicative of the periods of time when a reviewer should be reviewing a surgical procedure, reducing the amount of time required from various human reviewers. As provided in 2602B, and 2604B the machine-learning approaches may be utilized to reduce the amount of time the surgical procedure should be reviewed (as indicated by the reduction of 2602B relative to 2602A into staggered portions of time).
For example, instruction sets may be utilized to generate notifications and/or other representations of when a particular surgical record may require validation. In some embodiments, the determinations of when the surgical record should require validation by a reviewer may be established through filtering the timeframe to include portions where the potential likelihood of a clinical event as determined by the perception engine 2000 is greater than a particular threshold (e.g., 20%, 40%, 60%, 80%, 85%, 90%, 100%, or percentages in between). Accordingly, the cost and expense allocated to review may be better focused on time periods in which a potential error may have occurred.
In some embodiments, confidence scores and/or levels may be indicated through visual means and/or representations, such as the use of colors to indicate differing confidence levels. For example, red may be indicative of higher levels of confidence about a clinical event, orange may be indicative of medium levels of confidence, and green may be indicative of low level of confidence.
Example to Predict Locations of Surgical Error in Timeline
In some embodiments, Matlab™ code may be utilized to predict locations of surgical error. While other types of code (e.g., Python™, LISP™, C™, Javan™, Visual Basic™) may be utilized, the following is provided as an illustrative example. The Matlab™ code can be split into two parts. The first part of the code generates an Attribute-Relation File Format (.arff) file which can be used to generate a machine-learning algorithm. The second part of the code uses the machine learning algorithm generated in the first part to create a timeline with predicted locations of a surgical error to be provided to surgical raters to assist them with their ratings of various surgical procedures.
Generating an .arff file: The following steps are taken by the Matlab™ code to generate a .arff file: Matlab™ reads in an audio file, whose file name is specified by the user Matlab™ reads in a rated Timeline associated with the audio file, with technical error annotations specific to the audio file. Using the rated timeline, the Matlab™ code finds a timestamp of a surgical error, and cuts out an audio clip surrounding that timestamp from the larger audio.
The Matlab code extracts audio features from the smaller audio clip. The Matlab™ code prints the data about the audio clip into the .arff file. The Matlab™ code also labels the audio “yes”, as in “yes, there is a surgical error in this audio clip”. The Matlab™ code takes an audio clip from a random point in the audio not close to a surgical error annotation, and repeats steps 4-5 for that audio clip. The audio clip is labelled “no” since this audio clip has no surgical error in it. The Matlab code repeats the steps for all rated surgical errors and all surgical audios specified by the user.
The end result is an .arff file with a library of audio clips both associated with and not associated with a surgical error, with the numerical values of the audio features for each audio clip.
Generating a machine learning algorithm: The .arff file may be used by the hardware engine configured to generate decision trees to generate an Alternating Decision Tree (ADTree). This ADTree is converted from the output into a Matlab function, which takes the features of an audio clip as inputs, and outputs a prediction: whether the audio clip has a surgical error associated with it or not.
Generating a timeline with predicted error timestamps: Using the ADTree, Matlab™ then generates a file with annotations where the ADTree predicts there might be a surgical error. Matlab™ may be configured to utilize the following steps to do so:
Matlab™ reads in an unrated surgical audio file and the empty studiocode .TLCodes file associated with the audio. Matlab™ extracts relevant data, such as the video file location, from the empty studiocode .TLCodes file. Matlab segments the audio file into smaller pieces, whose length is specified by the user. For each audio segment, Matlab extracts features for that audio and runs the data through the alternating decision tree. Matlab keeps track of the timestamps where the ADTree predicts that a surgical error has occurred. Using the information from step 2 and step 4, Matlab™ writes the data into a pre-built .TLCodes template file. The end result is a studiocode .TLCodes file that, when opened, will produce the video associated with the audio file, as well as a timeline with locations of predicted errors.
This can be used by surgical raters to assist them with their surgical ratings.
A user interface may be provided where a user can specify an audio filename, TLCodes filename, TLCodes template filename, desired audio sample length, desired output filename, and a desired bias towards predicting errors, among others. After this, the user can click the “Create prediction timeline” button and a predictive timeline will generate in the same folder as the surgical audio.
Educational Strategies Generated Using the Black Box Data
Embodiments described herein may implement educational interventions based on OR black box performance analysis. For example, embodiments may provide training solutions or provide output data files that may be used to generate training solutions.
The data obtained from the systematic analysis of operative procedures may provide insight into the complex processes within the healthcare system, allow assessment of performance on an individual and team level, and evaluate human interactions with modern technology. Furthermore, this data can be used to determine specific individual and team performance deficiencies, hazard zones within procedures as well as characterize the cascade of events that result in “near misses” or adverse patient outcomes. This information may deliver critical knowledge content required to tailor effective educational interventions based on real life observations rather than hypothetical scenarios used in current training. This concept, grounded in theory of experiential learning may be used to create generalizable educational strategies that can be packaged and delivered to sites that do not have access to their own real-life data.
All training interventions may be tested using rigorous research methodology to generate a set of validated training solutions rooted in real observation.
The educational interventions may employ diverse instructional strategies such as team debriefing, individual and team coaching, error awareness and mitigation training, behavior modeling and warm-up simulation training.
Embodiments described herein may provide identification of root-causes of adverse outcomes and design of training scenarios. By way of example, the cause of adverse patient outcomes may remain elusive as they are frequently multifactorial and based on retrospective analysis. Embodiments described herein with black box generated data may allow analysis of prospectively documented adverse outcomes. Patterns of recurrent problems may be identified, characterized and used to generate a set of scenarios based on real experiences. This knowledge may be relevant to all OR teams involved in patient treatment in similar clinical contexts. The educational content may be compiled and delivered to information sheets, textbooks, e-learning software, virtual-reality simulation tools and software as well as integrated into SOPs at an institutional level.
Beyond summarizing common or significant root-causes of adverse outcomes, these scenarios may be used to generate software packages for full-scale simulations in virtual OR's. The variables can be programmed into the simulation software and thus be packaged, commercialized and exported to educational institutions worldwide.
Embodiments described herein may provide technical analysis to determine error frequencies, distribution and hazard zones. For example, the end-user of this data may be practicing physicians/surgeons and trainees. Mapping procedure complexity and identifying potential hazard zones can be used to create educational strategies targeted directly at these steps. Instructional strategies such as deliberate practice can then be used to train surgeons to be better prepared for these steps and thus minimize the risk of adverse events. Informing surgeons about complex or hazardous steps also enables the design of SOPs (such as in aviation for example with the “sterile” cockpit concept during takeoff and landing), to limit distractions during these sensitive steps (no irrelevant conversation, minimize room traffic, reduce overall noise).
Embodiments described herein may provide identification of beneficial and detrimental team interactions, and design and validation of simulated team training scenarios.
The functioning of the team may be influenced by non-technical skills such as communication. Non-technical skills have also been linked to patient outcome. Therefore, recognition of specific behavior patterns within teams that are either beneficial or detrimental to patient outcome is a step that may be required to subsequently fashion specific team training interventions and debriefing sessions. The core will thus use the data generated through the OR black box observations to identify specific patterns in non-technical performance of the teams. This information may serve as the basis for design specific team interventions using OR simulations, role-play and debriefing sessions. Recurrent themes that are identified as affecting team performance on an organizational level may be addressed by policy recommendations and the design of SOPs.
The end user of this data may be all inter-professional OR teams. Educational interventions derived from the black box data will be designed as a teaching package for inter-disciplinary team training. Behavior patterns identified to cause disruptions in organizational processes will be addressed by policy changes at local and regional level.
In an embodiment, there is provided a system for collecting and processing medical or surgical data. The system comprises a plurality of hardware units for collecting real-time medical or surgical data streams having a control interface coupled by a network to cameras, sensors, audio devices, and patient monitoring hardware, the real-time medical or surgical data streams relating to a real-time medical procedure within an operating or clinical site, wherein the real-time medical or surgical data streams include biometric data from healthcare providers during the medical procedure. The system also comprises an encoder with a network server for synchronizing and recording the real-time medical or surgical data streams to a common clock or timeline to generate a session container file. The network server is configured to control a multi-nodal perception engine to generate a protocol for data extraction from the session container file; process the data using the protocol to extract patterns for time-stamped clinical events within the session container file, each time-stamped clinical event associated with a confidence level; generate an interface indicator for a temporal sequence of the time-stamped clinical events within the session container file and error assessments, the interface indicator identifying each of the time-stamped clinical events and the associated confidence levels; and generate a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures with neural networks for modelling correlation of data for interference and feature extraction.
In an aspect, the system further comprises device middleware and hardware to establishes a secure reliable connection using a network infrastructure for communication with the encoder and the hardware units, the device middleware and hardware for translating, connecting, and formatting the real-time medical or surgical data streams received independently from the hardware units.
In an aspect, the device middleware and hardware implements data conformity and accurate synchronization for the real-time medical or surgical data streams using network protocols for clock synchronization between the hardware units to assist the encoder to generate the session container file.
In an aspect, the encoder and device middleware and hardware are operable to interface with third party devices to receive additional data feeds as part of the real-time medical or surgical data streams.
In an aspect, the system further comprises a central control station accessible using the control interface, the control station configured to control processing of the data streams in response to input control comprising play/pause, stop session, record session, move to session frame, split-display, recording status indicator, and log file.
In an aspect, the network infrastructure provides increased fail-over and redundancy for the real-time medical or surgical data streams from the hardware units.
In an aspect, the system further comprising a storage area network for storing data container files of the real-time medical or surgical data streams until scheduled transmission.
In an aspect, the encoder implements identity anonymization and encryption to the medical or surgical data.
In an aspect, the encoder processes the real-time medical or surgical data streams to generate measurement metrics relating to the medical procedure.
In an aspect, the real-time medical or surgical data streams correlates to a timeline, wherein the encoder detects events within the real-time medical or surgical data streams at corresponding times on the timeline, and tags and timestamps the session container file with the events, the timestamps corresponding to times on the timeline.
In an aspect, the system further comprises an intelligent dashboard interface for annotation and tagging of the synchronized medical or surgical data streams, wherein the intelligent dashboard may implement a viewer with playback viewing for reviewing content and interface controls for tagging content.
In an aspect, the intelligent dashboard is multi-dimensional in that the union of all dimension variables for the medical procedure may indicate a specific set of one or more applicable annotation dictionaries or coding templates.
In an aspect, example variables that may be used to determine the annotation and tagging dictionary may be: the type of medical procedure being performed, the aspect of the procedure that is being analyzed, the geographic area/region where the procedure is being performed.
In an aspect, the time-stamped clinical events within the session container file is stored with associated metadata for duration and frequency of each time-stamped clinical event.
In an aspect, the network server uses patterns for time-stamped clinical events within the session container file to identify and extract features from the session container file for correlation or spectral analysis based on temporal nature of the time-stamped clinical events within the session container file.
In an aspect, the network server identifies frequent temporal events as patterns leading to adverse events or errors in the timeline and develops predictive models to identify critical events during the real-time medical procedures.
In an aspect, the network server groups the time-stamped clinical events within the session container file into technical and non-technical events.
In an aspect, the interface indicator comprises an audio representation of the sequence of the time-stamped clinical events.
In an aspect, the network server configures the multi-nodal perception engine for filtering the time-stamped clinical events within the session container file using machine learning with feature extraction for event correlation using computational intelligence, the multi-nodal perception engine interfacing with distributed hardware units.
In an aspect, the network server generates the predictive data model by generating an Attribute-Relation File Format and the artificial intelligence network data structures and creates the temporal sequence of the time-stamped clinical events within the session container file, the temporal sequence relating to predicted locations of surgical error, wherein the interface indicator receives rating indicia for the error assessments of the time-stamped clinical events.
In an aspect, the network server implements post-processing of the time-stamped clinical events within the session container file for comparative processing for outcome links to generate hazard maps for the interface indicator.
In an aspect, the biometric data comprises heart rate variability measurements of the healthcare provider during the medical procedure.
In another embodiment, there is provided a multi-channel encoder for collecting, integrating, synchronizing and recording medical or surgical data streams onto a single interface with a common timeline or clock, the medical or surgical data streams including biometric data from healthcare providers during the medical procedure and received as independent real-time or live data streams from a plurality of hardware units, the encoder having a network server for scheduling transmission of session file containers for the recordings, the encoder processing the medical or surgical data streams to generate measurement metrics relating to a real-time medical procedure, the encoder configured to generate a protocol for data extraction from the session container file; process the data using the protocol to define patterns for time-stamped clinical events within the session container file; generate an interface indicator for a visual sequence of the time-stamped clinical events within the session container file and correspondence assessments; generate a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures.
In an aspect, the encoder generates as output a single session transport file using lossless compression operations.
In an aspect, the encoder detects completion of a recording of the data streams and securely encrypts the single transport file.
In an aspect, the encoder implements identity anonymization to the medical or surgical data.
In an aspect, the data streams comprising audio, video, text, metadata, quantitative, semi-quantitative, and data feeds.
In an aspect, the biometric data comprises heart rate variability measurements of the healthcare provider during the medical procedure.
In an aspect, the biometric data comprises at least one of: electrocardiography (ECG) data to assess the state of the autonomic nervous system and general cardiovascular data; electromyography (EMG) data to capture muscle fatigue; electroencephalography (EEG) to capture brain function; electrooculography (EOG) data to capture eye tracking; or intelligent shoe soles to assess the level of fatigue.
In another embodiment, there is provided a method for collecting and processing medical or surgical data. The method comprises receiving, at a multi-channel encoder, a plurality of live or real-time independent input feeds from one or more data capture devices located in an operating room or other patient intervention area, the input feeds relating to a live or real-time medical procedure, the input feeds including biometric data from healthcare providers during the medical procedure; synchronizing, by the encoder, the plurality of live independent input feeds onto a single interface with a common timeline or clock; recording the synchronized input feeds using a network server; generating, by the encoder, an output session file using the synchronized input feeds; transmitting the output session file using the network server; generating a protocol for data extraction from the session container file; processing the data using the protocol to define patterns for time-stamped clinical events within the session container file; generating an interface indicator for a visual sequence of the time-stamped clinical events within the session container file and correspondence assessments; and generating a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures.
In an aspect, the method further comprises processing the data streams for identity anonymization.
In an aspect, the method further comprises routing the data streams using a switch router to the encoder.
In an aspect, the biometric data comprises heart rate variability measurements of the healthcare provider during the medical procedure.
In an aspect, the biometric data comprises at least one of: electrocardiography (ECG) data to assess the state of the autonomic nervous system and general cardiovascular data; electromyography (EMG) data to capture muscle fatigue; electroencephalography (EEG) to capture brain function; electrooculography (EOG) data to capture eye tracking; or intelligent shoe soles to assess the level of fatigue.
In another embodiments, there is provided a cloud based system for collecting and processing medical or surgical data. The cloud based system comprises an encoder having a control interface for, in response to receiving a control command, triggering collection of real-time medical or surgical data streams by smart devices including cameras, sensors, audio devices, and patient monitoring hardware, the medical or surgical data including biometric data from healthcare providers during the medical procedure and relating to a real-time medical procedure within an operating or clinical site, the encoder for authenticating the smart devices, the smart devices synchronizing the real-time medical or surgical data streams by embedding timestamp markers within the real-time medical or surgical data streams, the timestamp markers generated by each smart device by a device clock; a media management hub server with middleware and hardware for translating, connecting, formatting, and recording the real-time medical or surgical data streams to generate session container files on network accessible storage devices; wireless network infrastructure to provide a secure network connection between the encoder, the smart devices and the media management hub server for communication of the real-time medical or surgical data streams; a central content server for storing and distributing the session container files and providing a two-way communication interface for the media management hub to implement a file transfer handshake for the session container files. The central content server is configured to: generate a protocol for data extraction from the session container file; process the data using the protocol to define patterns for time-stamped clinical events within the session container file; generate an interface indicator for a visual sequence of the time-stamped clinical events within the session container file and correspondence assessments; generate a predictive data model for refining protocol generation using support vector machines or artificial intelligence network data structures; and switching or gateway hardware to transmit the session container files from the media management hub to the central content server.
In an aspect, the media management hub server broadcasts clock data to the smart devices for synchronization of the device clocks.
In an aspect, the encoder provides a user interface to receive the control command and display real-time visual representations of the medical or surgical data.
In an aspect, the media management hub server aggregates, packages, compresses and encrypts the real-time data streams to generate the session container files.
In an aspect, the media management hub server manages the smart devices based on location, schedule, zone and requirements.
In an aspect, the media management hub server receives operating status data from the smart devices to generate a management interface with a visual representation of the operating status data for the smart devices, the operating status data including online, offline, running capture, and on-board storage.
In an aspect, the media management hub server processes the operating status data to detect smart devices operating outside of normal conditions and in response generating an alert notification of the detected smart devices operating outside of normal conditions.
In an aspect, the media management hub server implements a device communication interface for the smart devices to implement a device data transfer handshake for the real-time medical or surgical data streams.
In an aspect, the media management hub server authenticates the smart devices.
In an aspect, the cloud based system further comprises a computational intelligence platform for receiving the session container files to construct an analytics model to identify clinical factors within the session container files for predictions, costs and safety hazards, the analytics model providing a network for extracting features, correlations and event behaviour from the session container files that involve multivariable data sets with time-variant parameters.
In an aspect, the cloud based system further comprises a training or education server to receive the session container files, process the session container files to identify root causes of adverse patient outcomes and generate a training interface to communicate training data using the identified root causes and the session container files.
In an aspect, the smart devices include motion tracking devices for markerless motion tracking of objects within the operating or clinical site, the system further comprising a processor configured to convert captured motion data from the motion tracking devices into data structures identifying human factors, workflow design and chain-of-events.
In an aspect, the biometric data comprises at least one of: electrocardiography (ECG) data to assess the state of the autonomic nervous system and general cardiovascular data; electromyography (EMG) data to capture muscle fatigue; electroencephalography (EEG) to capture brain function; electrooculography (EOG) data to capture eye tracking; or intelligent shoe soles to assess the level of fatigue.
In an aspect, the biometric data comprises heart rate variability measurements of the healthcare provider during the medical procedure.
In an aspect, the biometric data comprises at least one of: electrocardiography (ECG) data to assess the state of the autonomic nervous system and general cardiovascular data; electromyography (EMG) data to capture muscle fatigue; electroencephalography (EEG) to capture brain function; electrooculography (EOG) data to capture eye tracking; or intelligent shoe soles to assess the level of fatigue.
The system 3200 is coupled to or integrated with hardware unit 20 or encoder 22 such that system 3200 is adapted to control or modify recording operations at a local server level (e.g., a computing node coupled to the recording devices of a particular operating room or location). Recorded information is provided into a data storage 3202, which can, for example, track biometric data obtained from biometric sensor 3290, which in some embodiments can be a biometrically enabled/sensorized shirt that captures ECG data. The system 3200 is adapted to operate in conjunction with the hardware unit 20 to generate an output data structure, such as a session container file (coupled, for example, with data sets representing annotations) from data streams captured during a medical procedure, which can be provided across network 3250 to a perception engine 2000 for alert generation using an alert module 3214 (e.g., software module, hardware alarm module).
The system 3200 includes a computer processor operating in conjunction with computer memory and a non-transitory computer readable storage medium. The system 3200 receives a biometric data stream from the biometric sensor 3290, which in this example, is coupled to a body of a healthcare practitioner, and captures electrocardiography (ECG) data.
The ECG data can be raw data, which can then be processed to obtain heart rate information, for example, as exemplified through tracked peak amplitudes (e.g., t=0, t=703 ms, t=1480 ms) relating to specific heartbeats of the individual. The ECG data, in some embodiments, is a digitized waveform having discrete values associated to voltage readings at particular timestamps (e.g., a reading every millisecond).
These voltage readings are obtained from electrodes of the sensor array that can be embedded, for example, in a smart shirt (e.g., improved scrubs). The electrodes can be coupled to various positions on a chest or torso of the practitioner, and multiple electrodes can be used together such that a signal is obtained from combining the various subsignals (e.g., useful where the shirt may not always be in good adherence with the body). The sensor array can communicate with the black box recorder system through wireless or wired communication mechanisms, such as Bluetooth, Zigbee, among others.
These tracked peak amplitudes can be used to obtain time deltas between biometric events extracted from the raw data, such as heart beats (e.g., 700 ms, 900 ms, 800 ms, 1100 ms, or 700 ms, 705 ms, 701 ms, 702 ms) represented by measuring frequency of specific data artifacts, such as high impulse spikes (e.g., the patterns characteristic of heart beats).
In this example, the time deltas are then utilized to obtain the HRV data values. For a given practitioner, a baseline HRV can be established by HRV extraction engine 3204 based on measurements tracked over a period of time by the biometric sensor 3290 during, for example, a duration of time designated as rest or normal. The table of
A biometric abnormality detection engine 3206 is configured to process the HRV values to identify a level of variance from the baseline HRV. In some embodiments, the level of variance can be measured in standard deviations, and where a pre-defined threshold of standard deviations is breached, a Boolean flag or other type of programmatic data object is toggled on a time-synchronized data structure indicative of a start of a duration of time related to an abnormal reading. This flag may be maintained until the deviant HRV is no longer detected, at which case, it is toggled off by the biometric abnormality detection engine 3206. In this example, a baseline HRV is 37, and a probability distribution function is applied to the data to identify a standard deviation (e.g., 0.7). In some embodiments, a normal distribution function is applied, but other probability distribution functions are also contemplated.
In this example, a SDNN of 34.4 is identified and compared against the baseline of 37. The difference is 2.6, which is 3.71 standard deviations (2.6/0.7). Accordingly, a flag is set toggled by changing, for example, a value having the variable name isAbnormalBiometric from FALSE to TRUE for the duration of time. In a variation, the duration of time associated with the deviant HRV is represented using a data object for each duration of time indicative of a start timestamp, and a stop timestamp (e.g., a tuple {9:00:01 AM, 9:14:56 AM}).
The data object can be encapsulated by the metadata tag generation engine 3208 and tagged to a particular procedure. While the “isAbnormalBiometric” flag can be used to generate alerts, this approach can be suboptimal as it can yield too many false positive values, especially for particularly stressful procedures where stress is normal (e.g., a rare, difficult, intensive, complex procedure). Accordingly, as described herein, the flag can instead be combined with modifications of the black box recordal mechanism for improving accuracy of downstream machine learning through a limited duration increase in resource allocation, and alerts can be generated only when a prediction data set (e.g., logits) indicates a predicted probability of an adverse event or error greater than a particular threshold or within a particular confidence band of occurrence.
Responsive to the actuation of a flag associated with a deviant HRV, the recorder control modification engine 3210 can be configured to temporarily modify recorder aspects to automatically improve a quality level of recordings generated by the hardware unit 20, albeit at the cost of temporarily consuming more computing and recording resources. Modifications, as described herein, can include increases in resolution, actuations of more recording devices, etc. Additional bandwidth resources may be acquired as well in advance of an expected increase in session container size for a stream to the perception engine 2000.
In some embodiments, modified encoding includes modifying a de-identification approach or other types of encoding transformations during the period of time. For example, a more resource intensive but more fidelity preserving approach can be used for de-identification for the limited duration of time.
The hardware unit 20 (e.g., a local recording controller circuit having a local processor) can be adapted to monitor an outgoing data feed or a data stream from the biometric sensors. Logical triggers in the form of automated logic rules can be established to switch from a first mode of recording to a second mode of recording upon detection of an analog or a digital signal representative of an abnormality-related duration of time. More, other modes are possible. The first mode may be a normal operation mode, where computer processing or bandwidth resources are maintained at normal levels, or, in constrained examples, maintained at low levels. One or more recording devices may be left off during this time. The second mode is a more intensive mode, where computational processing resources can be allocated or automatically instructed to operate at a high level (e.g., higher resolution, less compression, higher polling rate). When modes are switched, an instruction data set can be transmitted as a control signal for modifying the various aspects. Additional cameras or recording devices can be switched on/off, etc.
For example, upon a transition from mode 1 to mode 2, a control signal indicates that a wide angle camera should be switched on at 4K resolution, an overhead lamp camera records at 1080p instead of 320p, and a microphone should record full feature WAV files instead of highly compressed MP3 files. Other encoding aspects may also change, for example, a transition may switch a de-identification from an easier area “blur” type distortion to a more computationally intensive “probabilistic pixel cloud” obfuscation of the practitioner to preserve more features for downstream analysis. As the filesize is expected to increase, for example, from 1.2 Mb/s to 55 Mb/s, corresponding bandwidth resources can be allocated or dedicated. Similarly, in mode 2, the biometric data itself may be recorded to act as an additional input channel, but only during the limited duration of time of the abnormal characteristic.
In some embodiments, a machine learning resource control engine 3212 operates to request additional computational resources allocated to the perception engine 2000 to increase a speed at which perception engine 2000 is able to output one or more predictions through processing the increased feature set through the machine learning data model architectures of perception engine 2000 such that alerts can be generated in real-time or temporally proximate enough (e.g., within five minutes, half an hour, or a day) such that another practitioner is able to follow up or intervene if necessary. If the output prediction sets indicate a prediction greater than a particular probability threshold, an alert engine 3214 can be utilized to generate the alert to the second practitioner.
The alerts do not necessarily need to be real or in near real time, for example, in a variant embodiment, alerts can be generated on a batch processing event level during times of greater availability of computing resources (e.g., run nightly, weekly, monthly), and the alerts may be associated with post-operative outcome aspects. For example, the machine learning model may be tasked with tracking instruments (e.g., string, sponges) inadvertently left behind in individuals or poor technique (e.g., loose sutures, poorly adhering staples), and the alert may be generated when the machine learning model notes that there is a significant post-operative risk (e.g., for individuals having wider cross-sections and one or more comorbidities, it may track that the poor suturing may lead to an unacceptably high risk of a post-operative complication).
Over a period of training, the machine learning models can adapt to improve accuracy of analysis through supervised learning. Accordingly, the perception engine 2000 does not necessarily need to be hardcoded to observe specific combinations of events, rather, over time, through optimizing an objective function, the perception engine 2000 can generate predictions by processing the input features through a set of interconnected nodes representing a policy function through a set of refined interconnections (e.g., weights, filters) arranged in one or more layers.
In some embodiments, the additional computing resources are utilized by perception engine 2000 to conduct additional analysis using additional machine learning data models specifically adapted for use with the additional biometric data sets. For example, an expanded set of models can be utilized where the raw, processed, or semi-processed ECG data values are further included as an additional expanded input beyond the audio and video feature sets.
However, to conserve resources, the data channels associated with raw, processed, or semi-processed ECG data values are only provided to the perception engine 2000 during (or temporally proximate, such as 5 seconds before and after the triggering/release of the event condition) of the abnormality-related duration of time as indicated by the metadata tags or flagged time periods.
Embodiments described herein may contribute to improvements over current and/or previous designs. For example, embodiments described herein may provide scalability. Additional devices can be added to the configuration without excessive and costly hardware and cabling. As another example, embodiments described herein may provide optimization. They may be an improved ability to address varied physical spaces and add additional capture zones for wider range of event chains. As a further example, embodiments described herein may provide increased content with a greater ability to add additional data types for richer content. As an additional example, embodiments described herein may provide improved synchronization for devices with a reduced reliance on expensive hardware encoders, increased accuracy, and reduced exposure to latency. Embodiments described herein may provide greater leverage of general purpose computing equipment and reduced overall platform cost.
The embodiments of the devices, systems and methods described herein may be implemented in a combination of both hardware and software. These embodiments may be implemented on programmable computers, each computer including at least one processor, a data storage system (including volatile memory or non-volatile memory or other data storage elements or a combination thereof), and at least one communication interface.
Program code is applied to input data to perform the functions described herein and to generate output information. The output information is applied to one or more output devices. In some embodiments, the communication interface may be a network communication interface. In embodiments in which elements may be combined, the communication interface may be a software communication interface, such as those for inter-process communication. In still other embodiments, there may be a combination of communication interfaces implemented as hardware, software, and combination thereof.
Throughout the foregoing discussion, numerous references will be made regarding servers, routers, portals, platforms, or other systems formed from computing device hardware. The computing devices may have at least one processor configured to execute software instructions stored on a computer readable tangible, non-transitory medium. For example, a server can include one or more computers operating as a web server, database server, or other type of computer server in a manner to fulfill described roles, responsibilities, or functions.
The description provides many example embodiments. Although each embodiment represents a single combination of inventive elements, other examples may include all possible combinations of the disclosed elements. Thus if one embodiment comprises elements A, B, and C, and a second embodiment comprises elements B and D, other remaining combinations of A, B, C, or D, may also be used.
The term “connected” or “coupled to” may include both direct coupling (in which two elements that are coupled to each other contact each other) and indirect coupling (in which at least one additional element is located between the two elements).
The technical solution of embodiments may be in the form of a software product. The software product may be stored in a non-volatile or non-transitory storage medium, which can be a compact disk read-only memory (CD-ROM), a USB flash disk, or a removable hard disk. The software product includes a number of instructions that enable a computer device (personal computer, server, or network device) to execute the methods provided by the embodiments.
Although the embodiments have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein in different embodiments.
Moreover, the scope of the present application is not intended to be limited to the particular embodiments of the process, machine, manufacture, composition of matter, means, methods and steps described in the specification. As one of ordinary skill in the art will readily appreciate from the disclosure of the present invention, processes, machines, manufacture, compositions of matter, means, methods, or steps, presently existing or later to be developed, that perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein may be utilized.
As can be understood, the examples described above and illustrated are intended to be exemplary only.
This application is a non-provisional of, and claims all benefit, including priority to U.S. Application No. 62/907,001, filed Sep. 27, 2019, entitled “SYSTEM AND METHOD FOR BIOMETRIC DATA CAPTURE FOR EVENT PREDICTION”, incorporated herein by reference in its entirety. This application is a continuation-in-part of U.S. application Ser. No. 15/561,877, filed Sep. 26, 2017, entitled: “OPERATING ROOM BLACK-BOX DEVICE, SYSTEM, METHOD, AND COMPUTER READABLE MEDIUM FOR EVENT AND ERROR PREDICTION”, incorporated herein by reference in its entirety. U.S. application Ser. No. 15/561,877 is a U.S. National Stage of PCT Application No. PCT/CA2016/000081, entitled “Operating room black-box device, system, method and computer readable medium” filed Mar. 24, 2016, which claims priority from U.S. Application No. 62/138,647 dated Mar. 26, 2015 and to PCT Application No. PCT/CA2015/000504, entitled “Operating room black-box device, system, method and computer readable medium” filed on Sep. 23, 2015, which also claims priority to both U.S. Application No. 62/138,647 dated Mar. 26, 2015 and U.S. Application No. 62/054,057, filed Sep. 23, 2014. All of these applications are incorporated by reference in their entireties.
| Number | Name | Date | Kind |
|---|---|---|---|
| 9351652 | Dziubinski | May 2016 | B2 |
| 20070189544 | Rosenberg | Aug 2007 | A1 |
| 20070211758 | Aarnio | Sep 2007 | A1 |
| 20130006064 | Reiner | Jan 2013 | A1 |
| 20140309542 | Ahmed | Oct 2014 | A1 |
| 20170238859 | Sadowsky | Aug 2017 | A1 |
| 20180247024 | Divine | Aug 2018 | A1 |
| Number | Date | Country |
|---|---|---|
| WO-2015143067 | Sep 2015 | WO |
| Number | Date | Country | |
|---|---|---|---|
| 20210076966 A1 | Mar 2021 | US |
| Number | Date | Country | |
|---|---|---|---|
| 62907001 | Sep 2019 | US | |
| 62138647 | Mar 2015 | US | |
| 62054057 | Sep 2014 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | PCT/CA2015/000504 | Sep 2015 | WO |
| Child | 15561877 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 15561877 | US | |
| Child | 17035417 | US |
