Well-being, disease, and changes thereof are important aspects of health to monitor on a routine basis. Well-being is often tracked over time and repeated measurements taken to observe trend changes. Monitoring of disease often requires specialized penetrating sensing modalities requiring great expense to own and operate. Simplified and noninvasive sensing and monitoring technologies are often preferred to enable easier access to patient condition(s). One such sensing and monitoring technology that is difficult for human operators to utilize effectively to obtain information is a phonocardiography or listening (auscultation) through a stethoscope. The stethoscope has long been used by physicians as a sensing modality to diagnose disease and guide the evaluation of physical exam findings. Advancements in Artificial Intelligence (AI) in combination with auscultation devices have made it possible to resolve some conditions more effectively than human skills alone. Moreover, body sounds come from physical phenomena, the frequency components, of which, can be outside the range of human hearing (e.g., heart sound frequencies can extend below 20 Hz). Furthermore, body sounds are known to change with patient orientation. For example, auscultation of the heart of a patient in a reclined versus upright position can affect the sound because vertical changes in the pumping of blood volume changes the load on the heart. The same is true for auscultation sites at the carotid or femoral arteries. In addition, orientation of a patient also impacts the weight “above” organs such as when listening to organs in the abdomen. Knowing these differences is important to understanding the context of what is being heard and thus what physiological processes can be occurring. It would be beneficial if accumulated patients' data pertaining to multiple conditions and body orientations could be gathered through self-directed exams for analysis of sounds from a patient' body to automate diagnosis or to estimate health condition.
Embodiments of the subject invention pertain to devices and methods for providing guidance to a user for obtaining sounds from the body of a patient, collecting the sounds obtained, and analyzing the sounds to determine a physical condition of the patient. The devices and methods can include devices for acquiring and displaying phonocardiogram (PCG) data and methods for processing PCG data. More specifically, application of advanced processing methods, including artificial intelligence (AI), can enable display and analysis of sensor data obtained from a patient. Disclosed are all-in-one, networked, and cloud-based devices and a plurality of algorithms that make it possible to estimate the health of a patient and diagnose disease states.
Further enhanced, digital stethoscope can be used for motion sensing and obtaining other biometric data in addition to PCG data, which can be utilized together to segment exam sequences, resolve stethoscope positioning, and serve as additional inputs for AI processing. In one embodiment, infrared (IR) sensing is combined with digital stethoscope position mapping to resolve core body temperatures. Electrodes can be utilized for measuring impedance, conductance, and voltages further aiding the diagnostic perspective and features for AI processing. One or more devices can also be provided to allow for viewing, listening, haptics, or otherwise perceiving the information. The acquired information can then be processed by the AI algorithms to determine, for example, risks or likelihood of disease states, tracking and predicting disease trajectory (or state through time), cohorting subjects, estimating physiological parameters, estimating data quality, and guiding positioning of the digital stethoscope.
In order that a more precise understanding of the above recited invention can be obtained, a more particular description of the invention briefly described above will be rendered by reference to specific embodiments thereof that are illustrated in the appended drawings. The drawings presented herein may not be drawn to scale and any reference to dimensions in the drawings or the following description are specific to the embodiments disclosed. Any variations of these dimensions that will allow the subject invention to function for its intended purpose are considered to be within the scope of the subject invention. Thus, understanding that these drawings depict only typical embodiments of the invention and are not therefore to be considered as limiting in scope, the invention will be described and explained with additional specificity and detail through the use of the accompanying drawings in which:
The subject invention pertains to a digital stethoscope and related devices for implementing one or more algorithms and display schemes that aid in collecting and interpreting patient data acquired with the digital stethoscope. The term “stethoscope” refers to any device capable of being utilized as a directional vibration transducer. A vibration transducer refers to a device for detecting or sensing vibrations within or outside the range of human hearing, including, but not limited to, infrasound, ultrasound, and audible sound. In one embodiment, directionality is provided by a stethoscope bell, which guides sound (or vibrations) detected by the bell to a digitization device to be converted to an electrical signal. This can include directing sounds within or outside the range of human hearing, as well as sounds generated by the use of reflecting lasers or ultrasound off (or through) a surface to measure the vibrations, using a solid bell digital stethoscope rather than an air gap, or other contact or non-contact vibration measurements. Embodiments include a digital stethoscope and support systems for algorithms, display, and user interaction.
The term “patient” as used herein, describes an animal, including mammals, to which the devices and methods of the present invention can be applied and that can benefit from such application. This can include mammalian species such as, but not limited to, apes, chimpanzees, orangutans, humans, monkeys; domesticated animals (e.g., pets) such as dogs, cats, guinea pigs, hamsters; veterinary uses for large animals such as cattle, horses, goats, sheep; and, any wild or non-domesticated animal.
The term “user” as used herein refers to the operator or the person manipulating the digital stethoscope device on the body of a patent. The user can be the patient, another person manipulating the stethoscope device on the body of a patient, or some combination thereof.
One embodiment of a digital stethoscope 100 has a membrane 111 backed internally by a bell 110 and terminated with a transducer 113 as shown in
Furthermore, the digital stethoscope 100 can include any of a variety of sensors for detecting and transmitting the physical status of a patent or other patient data. In one embodiment, shown in
The digital stethoscope 100 can further include one or more accelerometers 120 and/or gyroscopes 140, which are sensors to detect and measure motion of the digital stethoscope 100. Motion can be integrated over time to estimate changes in the position of the device. In one embodiment, the digital stethoscope has one or more accelerometers to measure directional accelerations. In a further embodiment, the digital stethoscope has one or more gyroscopes to sense and/or measure rotational acceleration. In a still further embodiment, the digital stethoscope has one or more compasses to determine a world aligned absolute orientation without differentiating. The additional information from these positional devices can be utilized by the algorithms of the system to enhance performance and results. Understanding the orientation of the digital stethoscope can aid in determining the orientation or posture of the patient. Orientation or posture of the patient can be important to the interpretation of the sounds obtained by the digital stethoscope.
This system 200 of sensing, listening, processing, and/or visualization, as shown, for example, in
In one embodiment of the system 200, as shown in
Embodiments of the subject invention are particularly advantageous in the fields of medical and veterinary medicine, but can be beneficial in other fields where auditory signals are indicative of various conditions, such as automotive diagnostics. In the medical context, many auscultation sites can be utilized throughout the human body 301, examples of which are shown in
In the context of (AI) models for decoding system state, be it valvular disease in a heart or engine valves not sealing in a car, having the greatest signal to noise can be optimal. This is not, however, a requirement of successful data regression to create a performant algorithm or model. In many cases, designing for suboptimal input can be important, such as if it is possible the device will be operated by an unskilled user. Moreover, suboptimality detection can be the goal of AI to guide a user to make better, if not optimal, observations such that the user, device, and algorithm acting in concert results in maximal system performance. As a specific example, the aortic valve is usually best heard in the right, second intercostal space 306, as shown in
The aforementioned use examples can be enhanced through data-driven application specific algorithms utilized by supplemental compute resources 207. Examples include artificial intelligence (AI), neural networks, regression, Markov models, Gaussian mixture models, support vector machines, random forest, and other machine learning (ML) algorithms and methods for the diagnosis, guidance, detection, screening, monitoring, risk analysis, and prognostication, which we henceforth refer to as artificial intelligence or machine learning models. In the medical context, these can be applied to any number of ailments or current health conditions of the body.
In the medical context, these machine learning models can be applied to any of a variety of ailments or health conditions of the body. In general, two kinds of data are collected to train machine learning models: cross-sectional and longitudinal. In the cross-sectional case, one set of inputs can be mapped to another set of outputs. In the context of disease, for example, a single, brief, 10 second, PCG input can be modeled to predict the heart failure severity or current physiological parameter, such as regurgitation. In a cross-sectional dataset that utilizes similar PCG input, a machine learning model can be regressed or trained to understand the relationship between the PCG signal (or features) and estimate a patient's current regurgitation causes or severity. In a longitudinal dataset, multiple time points are used, potentially collecting changes in PCG signals, severities of events, abnormal events, etc. Then portions of the sequence of information can be used to predict current, or more importantly, subsequent, events or outputs, such as, for example, the severity of heart failure. This provides a tool, wherein an algorithm can predict the time to clinical worsening or the clinical state at some point in the future.
One embodiment of the subject invention is a device implementing a heart age metric. Using heart sounds and age information of patients, a model can be trained to predict the patient's age from the heart sound. Subsequent sounds are estimated to mostly likely be from a patient of the determined specific age. This model relies on a wear concept of the heart as it ages and changes in vasculature, since the resistance of the circulatory system often affects how the heart works. This has health monitoring applications where individuals are interested in monitoring their heart due to hereditary conditions, for positive feedback related to cardiovascular exercise, or any number of other situations. Furthermore, the result can be simple for users to understand: naturally occurring changes are okay and everyone ages, yet there is common perception that health declines with age after 18 years old. Thus, being accelerated in age for adults has negative implications, and the inverse is also true. Age associated data pertaining to physiological changes are more readily acquired and can provide a wealth of information for training models, as opposed to models designed for specific disease pathologies.
This methodology can be extended to other body systems and organs, including one that uses a lung age metric. These datasets can be augmented with other (potentially user entered) data such as smoking history, which can greatly affect many organs. Addition of the information in a machine learning framework, allows for separate understanding of cohorts of healthy patients versus cohorts of patients with suboptimal health or conditions that can skew the input distribution toward pathologic expressions. Moreover, mortality information can be applied to normalize the concept of “age” given the life expectancies of each cohort. Historical and current behavioral data can be one form of additional input and training data, however, additional information comes in the form of other sensor measurements, too. For example, blood pressure can be separately measured and input into the system for more accurate diagnosis.
Furthermore, data can also be applicable directly to pathology detection, severity determination, and risk stratification. For example, patient record information, such as Coronary Artery Disease (CAD) state, can be regressed against the sounds sampled from the heart. Turbulence in arteries, changes in body sounds 201 due to wall motion abnormalities, and alterations in the pumping cycle duration and conduction paths along the heart wall can all affect the body sounds that are observed. The learned model can then be applied to undiagnosed individuals (
In one embodiment, shown in the flow chart in
The embodiment above illustrates the algorithms that can be utilized by the supplemental compute resource in the system 200. Furthermore, in a patient context, the method shown in
Datasets that comprise heart sound recordings from a group of patients at one or more points in time can be utilized to predict patient mortality 5 (five) (or N) years later. With a dataset of one sample per individual and mortality information providing a precise time frame after the recording, it is possible to build a model similar to the usage of the data case above that is regressed against future data providing a predictive output. However, a more flexible model that can be employed that uses multiple samples of the intended inputs and outputs such as, for example, a dataset containing multiple heart sound recordings obtained from annual health exams and the corresponding diagnosis dates. Using that dataset, a model can be regressed to map sequences of heart sounds and a query time point to an individual diagnosis state. Then, a whole heart sound history can be input to the algorithm and then queried for the time frame of interest. This enables a user to use the stethoscope system to track progress toward their future selves or understanding their rate of deterioration or the disease trajectory in general.
In one embodiment, this trajectory through time is processed and displayed for one subject relative to a population. This can be performed in many ways, for example using statistical measures of the signal, through latent AI model features, or by dimensionally reducing input, intermediate, or output values of the model. Plotting of such data can give a visual context that simplifies understanding of the patient condition and probable future. Over time, the user will be able to view the effect of any changes to the subject by interventions. Moreover, it can visually confirm that interventions are having a desired effect.
In another embodiment, an AI is trained to predict future interventions and optimum timing therefore. When the subject is monitored using the device, the processing results can show when future interventions may be needed. For example, stenosis in a catheter that increases over time can be monitored for progression and when intervention will be necessary. The digital stethoscope 100 can be used to listen to the flow inside the catheter. The sounds (potentially due to the turbulence of the fluid) are processed by AI to infer the current state of the catheter and, using the history of measurements and/or typical progressions, predict when the catheter's stenosis will exceed prescribed conditions. Then, necessary interventions can be scheduled in advance to remedy the upcoming conditions. There are possible applications for nearly every condition, including but not limited to valve surgery, mechanical valve replacements, fistulas, and peripheral artery disease.
AI decision attribution can build trust with the user 405, 406, result in effective use of time, enable the user to make accuracy decisions using prior knowledge, and automatically annotate data for future AI training. These aspects can also build tremendous value from the perspective of tool utility and compounding data richness. Attribution is a developing interest in the increasingly complex field of AI models. Here, AI attribution methodologies are additionally applied for the development of a novel disease process isolation and noise reduction algorithm and workflow (
In one embodiment, after acquisition 401, 601 and processing of the data 403, 405, 601-604 as shown in
Beyond demonstration of the signal of interest (shown via attribution), an enhanced filtering technique is disclosed to amplify the components of the signal relevant to diagnostic outputs. In one embodiment, shown in
The processed results are reported to the user or reviewer, as shown in step 608. They can be presented with each estimated likelihood in an easy-to-understand format. Elements which mark elevated likelihood are selectable for further inspection, as shown in step 609. Once selected additional information is provided with regard to the disease or condition of interest and a playable waveform presented, where regions are annotated or highlighted to direct attention to regions of the signal, as shown at step 610. Regions of the signal, a list of subsamples, or a scrubbable cursor preplaced at the best example, step 611, allow the user to select priority listening regions to playback the sounds, as shown at step 612.
During playback of a processed signal, the user can be presented with a slider or pre-configured options for noise reduction, signal isolation, and/or signal amplification levels. Mapping the AI rationale back to components of the original signal and providing an interface to the user to magnify the basis for determination or suppress the original signal's distracting components, creates learning opportunities, trust, and enables the user to decide whether to rely on the AI's results. These culminate in improved combined user plus AI system performance and relationship with the goal of achieving super-human performance through contextualization of results and sample collection. This is particularly the case with an expert user, however, it may not be the case with untrained users which rely solely on AI guidance.
The AI guidance system enables untrained and trained users to collect specific or multi-site data (i.e., from multiple specific recording sites). This system consists of several logical components: a position and quality inference component, a reference and historic data baseline component with a stopping method, and a positioning and repositioning display.
With respect to the position and quality inference component, one embodiment is a machine learning algorithm that simultaneously computes both measures, trained using ground truth position information and expert scored quality. The advantage of simultaneously estimating position and quality can be that they are linked due to a prior for the desired position of the sensor measurement. Expert rated quality can be based on the ability to identify sound subcomponents expected in a specific position. For example, the aortic valve sounds when auscultated at the aortic site, if instead the mitral site were auscultated other sounds may be clear but the aortic valve may not be. Trial data can be expert re-rated in the context of several desired sounds or position contexts. The expected clarity can be, but is not limited to, being dependent only on position, several other factors are at play. For example, the strength of the heart, the amount and kind of tissue between the sound source and the surface of the body, external material between the skin/stethoscope interface (i.e., clothes, hair, oils, undergarments, etc.). These increase sound attenuation while others increase ambient noise. Ambient noise can come from speech (internal or external to the subject), background nuisances, or purposeful sounds. Quality can then be a measure of this overall clarity of target sources.
The reference and historic data baseline component determines the expected achievable quality. This expected achievable standard can be specific to each user or subject as the system learns specific information about the user, such as, for example, BMI, specificity with which they relocalize (how good they are at targeting and retargeting a spot with guidance), expected achievable background noise, and expected sound levels. Embodiments of one component of the subject invention can determine user feedback like positioning movements, alterations in the environmental noise, alterations in the device contact, and that user has achieved the AI's expectations and can complete the acquisition procedure. This component can use the results from the position and quality estimator such as the position, the quality, foreground and background sound levels, detectability of characteristic components, and others. In addition, the statistical information of these outputs from prior acquisition attempts by this user on this subject, statistical information of these outputs from cohort users or subjects, and statistical information of these outputs without prior information can also be employed in determination of the stopping condition.
With reference to
The nature and extent of repositioning guidance (for resampling or new samples) can be important in enabling the user to acquire the intended data source. In
In one embodiment, a face-on view can be shown for guidance directions. In a further embodiment, additional viewpoint specific views can also be given to aid the understanding of the guidance more rapidly. For example, if the user is the patient, then an additional egocentric view can be presented. In the case of a heart exam, an apparent over the shoulder view or a view looking down at oneself can increase the patient's left/right motion determination, as such view does not need to be reversed from the face-on view. Users that are the patient can occur in many scenarios. For example, when the user is a remote patient or an isolated patient. If the user is not the patient, then the face-on view is representative of the exact experience of the motion required. Supplementary, side, or back facing views would be appropriate when a user is auscultating the back of the patient or when using alternative stances. The direct queues 805 may not be suitable for user viewport-based views, as the user's view can be directly in line with the three-dimensional ray formed by the guidance. However, three dimensionally rendered animated guidance 804 showing a device relocating by moving toward or away from the viewport can be understandable from depth queues.
The captured view of the user can further be used to determine an estimate of the positioning of the stethoscope and the pose of the patient. The relative information can be used to determine current or desired chest placement. Another method is to directly estimate the relative chest positioning using a deep neural network. These methods can be implemented with feature extraction and matching algorithms, or through deep neural networks trained to make direct estimates. For example, marker tracking on a device has pre-implemented solutions like ‘apriltags,’ as well as off-the-shelf pose estimation models like ‘movenet’ for human bodies. Furthermore, image-based implementation of the stethoscope position estimation can be used to enhance position estimation from sounds of interest, as well as on their own. In general, the estimation of the position of the user enables drawing on screen the desired position 1303 in an augmented view of the user on the screen 1402.
Pose estimation of the user can be used as a basis for user feedback or initialization. For example, the aforementioned pose estimation model finds nose, eye, ear, shoulder, elbow, hand, waist, knee, and foot of the user in the frame. By simple comparison of position values in the frame, one can determine if the user is: facing the camera, vertical with respect to the camera, a good distance away from the camera for analysis, and positioned well in the frame. With satisfactory conditions, it is possible to define a relative coordinate system using the shoulders to estimate a position where various heart sounds can be sensed. In one embodiment, a shoulder to shoulder line segment 1306 and it's perpendicular bisector line segment 1308 can be used to define a relative coordinate system that scales with the width of the user's shoulders in the image frame. In relative terms, predefined coordinates for auscultation sites 306-309 are scaled. Then a marker 1303 is drawn on the image to display to the user an initial position to begin an exam. Feature tracking can be used to observe the placement of the stethoscope, which can be limited to the region of the hand (from the pose estimation) to speed processing. Even if the stethoscope is not directly visible (for example, if obscured by the hand) the pose of the hand can be used estimate positioning or provide additional evidence for positioning if used in conjunction with sound features.
Certain embodiments of the subject invention pertain to a method of utilizing a stethoscope 100 having an internal accelerometer 120 to resolve potential physical positioning or orientations of a patient, as shown, for example, in
In a further embodiment, the invention is extended to multi-site measurements and analysis. The possible orientations of the body regions can be further constrained by repeating the procedure at a site that is orthogonal or on a different geometric plane, and/or by additional AI processing that detects the most likely positions from sound. The accelerometer 120 of the digital stethoscope 100 can provide both the orientation at each site, and the differences in position and/or orientation between the two or more sites. An embodiment is shown in a flowchart in
Moreover, when monitoring multiple sites such that a previous position estimate is available, accelerometer data can be integrated to calculate an initial estimate for the position of subsequent auscultation sites. For example, if aortic valve sounds are determined to be present at one site, that can indicate proximity to the location of the aortic valve. Then, given accelerometer readings indicating motion of the distance and direction that separates the aortic valve and tricuspid valve, there are strong initial conditions for assuming that sounds being heard likely contain the tricuspid valve sounds. This can be further extended by using an estimated body size (through height and weight measurements, for example), to increase the accuracy or normalize across body sizes. Further, this combination can be used as a verification method of non-expert users in acquiring the correct data (or from the correct site). One embodiment of verification is a system in which the user is directed and places the digital stethoscope 100 on the patient. The system can estimate the position of the stethoscope based on sensor measurements. Then, the user can relocate the stethoscope to a new specific site and given real time feedback as to the correct positioning by using the expected change in the body sound source.
In a further embodiment, the accelerometer 120 is utilized to resolve the occurrence of relocations of the digital stethoscope 100 on a patient. Distinct changes in motion occur during exams which are monitored to automatically segment any data acquired. During digital stethoscope recordings of a patient, users can move from site to site quickly, dwelling at each site for only a few seconds. During the dwell time, the stethoscope is preferably maintained in position with minimal movement. Users can minimize movement to reduce the noise and maintain near optimal positioning. In addition, when relocating or repositioning, users tend to lift, reposition, and place at the site. Adding an accelerometer enables monitoring of accelerations indicative of repositioning (or movement).
Segmentation and localization of the exam data stream using the accelerometer 120 and/or sound data can enable visuals and interactions. For example, playback of a data stream containing multiple recording sites can be started. In one embodiment, a user can listen to the continuous sound stream while a display highlights the current site source of the currently sounds, as the stethoscope is moved. In an alternative embodiment, multiple recordings can be arranged sequentially or “strung together” and played in exam sequence. In another embodiment can be extended to reordering of exam sequences allowing the recording in any sequence and playback in the preferred sequence by user. Furthermore, segmentation and localization can be further connected with a user interface where sites are selectable. The user can then select a site that corresponds to that portion of the recording and playback the applicable data segment. Adding classification to the segmentation and localization provides the ability to visually label each site as appropriate with detected conditions. For example, the aortic auscultation site can be labeled with valvular disease or high regurgitation while the Erb's point 308 (an auscultation site during a heart examination, located in the third intercostal space close to the sternum) can be labeled with wall motion abnormality. This visual map of health conditions can facilitate user understanding of the exam results.
The accelerometer 120 or other inertial measurement information can be directly integrated, estimating the motion of the stethoscope during recordings. Over short time periods, the estimated movements are typically accurate and can be made more accurate by compensating for drift that might be integrated during the recording when the device is known to be stationary at a site through other cues (for example, the sound data). These estimated changes in position can be used to constrain the optimization of recording positioning.
The accelerometer data can be further processed to determine events occurring in the streams and/or de-interlace source components. For example, the accelerometer data stream can contain overlapping and diverse sources such as motions to new sites, on-site adjustments, chest motion from cough, chest motion from breathing, tremors, body movements, breath holding, valsalva maneuver, etc. An AI model can be applied to the data sequence to find, classify, and separate the signal components due to each event. The AI can produce a signal for each component that is separated. The classifier labels event onsets and enables aggregation and display of exam statistics. The classifier can trigger further AI processing of the sounds, for example, of a cough, which have shown to contain important health information in conditions like pneumonia or COVID-19. The segmentation of accelerometer and/or sound data into each discrete event and/or repeating events can be labeled for training AI models. This pre-demarcation can reduce the time required to create labeled datasets and enables AI- and human-in-the-loop iteration for labeling the dataset where the human identifies examples, edge cases, and errors and the AI can label the bulk of the data from what it has learned so far. The processed events can also be functional or exam procedures, such as breath holding that serves multiple purposes such as modifying the volume of blood returning to the heart while also limiting lung or bronchial sounds. Downstream AI can then be triggered to process only the lower noise version of the heart sounds and/or with added context to the processing of said data.
Algorithms utilizing multisite or multistate data are fed by guidance directing the user to acquire data from each of the sites and/or in each subject state (i.e., orientation, stress level, cycle phase [time of day, time since last drug dosage, menstrual, and more], etc.). These algorithms can greatly expand upon the capabilities of single source algorithms. For example, data from multiple sites can be used to localize the source of signal components and, thus, imply the physical component generating the sounds. Further, multiple states of stress can be used to determine or grade the vitality of individual internal components. Not only does this help with prognostication, but can also reveal issues that can otherwise be hidden. As a non-limiting example, when auscultating a heart at rest it may sound functional, however the added stress of high intensity cardiovascular activity may reveal divergences from a normal sounding heart. As a further example, a user and patient can be guided to use the stethoscope at rest at the mitral auscultation site, then be directed to perform ten minutes of exercise and examine again before the heart begins to relax. Both recordings together contain more information about the state of the heart than either alone.
Embodiments of the subject invention can include multi-state measurements and analysis. As used herein, multi-state refers to more than one measurement of different body positions, activities, stimuli, and/or any other condition in which there is a change in body function or response. For example, recording from the same site (or multiple sites) in an upright position and a reclined position with these changes in state being automatically detected and the data automatically segmented and organized for automatic AI processing and classification using the prior knowledge of the multiple states of the subject. Moreover, the accelerometer can provide a direct measure of the states and any differences therein. The difference between upright and reclined subject measurements can vary between users and the absolute positioning of those states can also vary amongst users. Measuring this accurately captures the spectrum of changes and states possible which is important data for training and decision making driven by AI to maximize results. This is especially true in the context of a self-exam or inexperienced user where the subject cannot strictly adhere to exam guidance (for example, by slouching). Further examples of multi-state include leaning from one side to another, before, during and after surprising or emotional stimuli, or before, during and after exercise.
In a further embodiment, the accelerometer 120 in the digital stethoscope 100 is utilized to measure activity of the subject during at least one of prior to, during, and after a sound recording. This can be achieved with a carry-on person device or as a wearable device. Activity can be a predictor for many health concerns. If an embodiment of the device is carried on or near the patient, for example in a pocket, it can be used to monitor the heart or lungs, and AI can be trained to account for the varied effect of prior activity on heart output. This can increase sensitivity of the AI to abnormalities because of the increased ability to compensate for patient context. In one embodiment, prior activity monitoring can be conducted by, but is not limited to, the stethoscope being carried with the subject. In another embodiment, the carry-on device used to monitor a patient is a wearable device attached to the patient using a strap, adhesive, or some such other means to enable relative station keeping on the patient. The accelerometer can also be embedded, attached, or otherwise arranged on the patient or clothing, a separate activity monitor wirelessly connected to the stethoscope apparatus or an application host mobile device, or in an application host mobile device.
In another embodiment, the accelerometer 120 is utilized to measure motion or vibrations larger or longer in wavelength than the digital stethoscope transducer can measure. One example (depending on the exact digital stethoscope sensor used) is the motion of the chest during inhalation and exhalation. The stethoscope can rise and fall or deflect in and out with each breath of a subject. This pattern of motion can be tracked and used to estimate the relative depths of breaths, the phase of breaths at any moment, and synchronization with other body functions. Many body functions are modulated by nearby pressures (a blood volume effect), for example, there is a breathing effect on heart function. Inhalation creates negative pressure, through movement of the diaphragm and rib cage, drawing more blood volume into the thorax and increasing blood flow to the right side of the heart. In some disease states, breathing is supported by a machine using positive pressure ventilation. Each of these modes has the potential to change pressure in the thorax and alter the volume of blood returning to the right heart from the body. This alteration in blood volume on the right side of the heart is not present in the left side of the heart for several heart beats. The difference in volume can be detected using the PCG. Therefore, tracking breaths can provide valuable context to a heart exam. Moreover, the increased pressure from each breath can act as a multi-state driver for the heart function. Then each heartbeat can be analyzed in the context of varied stress as applied by the diaphragm or support machine. This phenomenon is not limited to the breath—pulse relationship, for example the same framework can be applied to gastrointestinal function.
With the understanding of the pressure and blood volume mediated relationship between the breaths and heartbeats, an estimate can be made as to the volume responsiveness of a patient. That is, if the blood volume preload at the heart increases, the increase in stroke volume (or cardiac output) can be determined. This is possible by processing sounds of blood flow velocity and duration in the heart under varied pressures induced by the diaphragm or support machine. In addition, certain disease states will alter these measurements and can be diagnostic of the disease state. Hypovolemia secondary to dehydration or hemorrhage, for example, could be indicated by alteration of these pressures. Systolic dysfunction is also potentially detectable through monitoring of decreased cardiac output.
In a still further embodiment, an IR sensor 114 of the digital stethoscope 100 is utilized with the sound data and accelerometer data to increase the accuracy of disease detection AI, particularly for conditions known to alter body temperature. The infrared sensor can measure the temperature at each recording site and/or the rate of change of the temperature of a digital stethoscope 100 component that is in contact with the patient. These data can be used to estimate the surface temperature of the patient. When used with the digital stethoscope detected sounds, AI processing accuracy can be improved for pulmonary and cardiac conditions, such as, for example, pneumonia. An increase in body temperature is expected when there is an active infection of the body. The inclusion of temperature with metrics for breathing, coughing, heart output, etc. can build a holistic picture for analysis to determine infection, shock, or other temperature modifying conditions.
Furthermore, combining the temperature readings, obtained with the IR sensor 114, with the estimated position of each reading, can enable more accurate inference of core temperature as external body locations vary with respect to their difference relative to body core temperature. Thus, incorporation of the accelerometer data with the phonocardiographic (PCG) data, as shown, for example, in
In another embodiment, the digital stethoscope can have a strain gauge 130 sensor that is utilized on the patient contact side of the digital stethoscope 100. One placement for a strain gauge can be in or on the membrane of the digital stethoscope. Another placement option can be in the bell of the digital stethoscope such that it is protected from patient contact, yet is able to measure small deflections in the structure of the stethoscope when in contact with the patient. The strain gauge can measure one or more pressures applied to the membrane material giving a proxy for the forces applied to an auscultation site. Those forces can potentially indicate a need to displace body fat and, thus, can further indicate the amount of body fat at that location. In a further embodiment, the strain gauge measures the displacement of tissue edema. By way of a non-limiting example, when the digital stethoscope is pressed against the pretibial tissue, the pattern of pressure increase can be used to estimate the level of pitting edema present. Higher levels of edema are noted through smaller rebound pressure of the tissue.
In a further embodiment, a digital stethoscope 100 in combination with the strain gauge 130 can be used to press on the brachial artery until blood flow is obstructed. Then, slowly applying less pressure on the brachial artery, the Korotkoff sounds can be heard. As further pressure is released, the Korotkoff sounds cease. The AI processes the Korotkoff sounds to determine the beginning and end of the sounds and uses the strain gauge values to interpret the systolic and diastolic pressures. This combination can enable a one-device solution, which can eliminate using an external pressure cuff with a standard stethoscope.
A still further embodiment utilizes a combination of the digital stethoscope 100 with a plurality of electrodes 150 on a rim 110 of the digital stethoscope device 100, as shown, for example, in
Management of pulmonary artery pressure (PAP) via changes of medication in heart failure (HF) patients has been shown to reduce hospital readmissions. In one embodiment, shown in
Remote health examinations have become commonplace and often require patients to provide information and obtain personal health data utilizing unfamiliar medical equipment and device. The embodiments of the subject invention provide devices and methods that improve the comfort and efficiency of patient handling of the medical equipment or devices utilized for obtaining personal health data. Embodiments of the subject invention provide a digital stethoscope with one or more sensors that can detect sounds and conditions of a patient and can transmit the sounds and information to an AI. Advantageously, the AI can analyze the incoming sound information and determine a direction in which the stethoscope should be maneuvered on the patient body to obtain more accurate information from the patient. When the exam is complete, the AI can analyze the sound and other information received and compare against a cohort of other patients to determine if the sounds and information are indicative of a particular condition. The embodiments described herein are particularly advantageous for detecting and diagnosing heart conditions. When coupled with an application for guiding the patient movements of the digital stethoscope, the patient can more easily and accurately position the digital stethoscope.
The methods and processes of using at least one of the disclosed exam guidance, exam characterization, exam completion determination, attribution interface, and/or attribution-based filtering represent novel inventions from the state of the art. Their combination with a digital stethoscope and/or AI processing for specified use cases represent further inventions disclosed herein. Moreover, many said AI processing apparatus disclosed throughout (i.e., a device processing “heart age” or “lung age”) are novel in their outputs as well as multi-site and/or multi-state inputs. In addition, application of these systems to predict scheduling of interventions represents a new and non-obvious workflow for expert users.
All patents, patent applications, provisional applications, and other publications referred to or cited herein are incorporated by reference in their entirety, including all figures and tables, to the extent they are not inconsistent with the explicit teachings of this specification. Additionally, the entire contents of the references cited within the references cited herein are also entirely incorporated by reference.
The examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application.
The invention has been described herein in considerable detail, in order to comply with the Patent Statutes and to provide those skilled in the art with information needed to apply the novel principles, and to construct and use such specialized components as are required. However, the invention can be carried out by specifically different equipment and devices, and that various modifications, both as to equipment details and operating procedures can be effected without departing from the scope of the invention itself. Further, although the present invention has been described with reference to specific details of certain embodiments thereof and by examples disclosed herein, it is not intended that such details should be regarded as limitations upon the scope of the invention except as and to the extent that they are included in the accompanying claims.
This application claims the benefit of U.S. Provisional Ser. Nos. 63/320,240, filed Mar. 16, 2022 and 63/345,052, filed May 24, 2022, the disclosures of which are hereby incorporated by reference in their entirety, including all figures, tables and amino acid or nucleic acid sequences.
Number | Date | Country | |
---|---|---|---|
63320240 | Mar 2022 | US | |
63345052 | May 2022 | US |