The present invention relates to a physiological data analysis device and system, in particular to a device and system to collect, process and display physiological data of multiple types, gathered by sensing devices of different types or function and defined in various forms and descriptions, in order to find a correlation of a type of physiological data and specific physiological phenomenon.
Polysomnography (PSG) is the most commonly used standard inspection method in sleep medicine and the diagnosis of sleep related diseases such as sleep disorders, snoring, epilepsy, and sleep apnea. The inspection is usually carried out in a hospital ward. The patient must stay in the hospital, usually in the sleep center, and the doctor or sleep technician installs a variety of sensors on the patient to gather the sleep related physiological data throughout the night. The inspection results are displayed at intervals of, for example, every 30 seconds. Taking a 6-hour inspection as an example, 720 units of inspection results will be produced, which are then processed and provided to the doctor for diagnosis.
PSG needs the combination of multiple instruments to complete the inspection and provide for comprehensive diagnosis. Inspection items usually include:
As there are too many items to be inspected by PSG, the multiple inspection instruments attached to the patient do not only affects the patient's sleep, but also lead to inaccurate detection. In addition, the statistics and marking of the results are also quite labor-intensive. To solve this technical problem, the industry has proposed a variety of solutions that performs fewer types of inspection items, supplemented by software, to automatically mark inspection results. For example, for the diagnosis of sleep apnea, a simplified sleep physiology examination device was developed. The device only needs to measure nasal airflow, pulse, and blood oxygen concentration. The collected data can be interpreted by a machine to generate a sleep apnea test result similar to PSG, namely the sleep apnea index (Apnea-Hypopnea Index, AHI).
A paper by Sun et al. found out, after deep learning with a large amount of PSG data, that adding the value of abdominal tension to the ECG signal, it is possible to calculate a sleep staging result that is quite close to diagnosis results using brain waves. See Haoqi Sun et al., “Sleep Staging from Electrocardiography and Respiration with Deep Learning.” Dec. 21, 2019, Sleep 2020, https://academic.oup.com/sleep/article-abstract/43/7/zsz306/5682785.
Largan Health AI-Tech also performed machine learning on a large amount of PSG data and announced a sleep analysis software that uses ECG signals, only, and provides diagnosis results quite close to the sleep staging and apnea index by using PSG.
With the popularization of wearable devices, IoT sensing technology, and millimeter wave technology, many experts try to place more instruments on the subject, hoping to more accurately detect and predict certain physiological phenomena, and/or find out the cause, seek ways to improve health. However, instruments or measurement methods that are useful for the detection, prediction, or cause analysis of physiological phenomena have not been discovered because of the new technologies and new products. The accuracy of the measurement has not been improved accordingly, either.
The purpose of the present invention is to provide a novel multiple physiological data collection and analysis device, as a solution for multiple physiological data collection, manual marking, machine learning, training sampling, AI analysis and other processes, all in a single environment.
The objective of this invention is to provide a tool that is convenient for professionals to quickly find in the vast sea of data the types of physiological data that are correlated to specific physiological phenomena.
The present invention provides a multiple physiological data collection and analysis device that obtains/receives physiological data from various sensing devices and automatically classifies and stores the physiological data. After machine learning, certain types of physiological data correlated to specific physiological phenomena can be found and evaluated.
The present invention provides a multiple physiological data analysis system that can display different types of physiological data received from various sensing devices on the same display device according to the conditions set by the user. The present invention provides a convenient tool for researchers to discover a correlation of a type of physiological data and certain physiological phenomena.
The present invention provides a multiple physiological data collection and analysis device that provides useful evaluation tools to determine the correlation between specific types of physiological data and specific physiological phenomena.
To achieve the above objectives, the present invention provides a multiple physiological data collection and analysis device, which comprises:
a data upload device to provide a communication channel for communication link of a plurality of physiological data sensing devices or physiological data storage devices, to receive different types of physiological data from the plurality of physiological data sensing devices or physiological data storage devices;
a data storage device to provide a large memory space for storing various physiological data and result data of the physiological data processed in the multiple physiological data collection and analysis device;
a data editing device to provide a human-machine interface for users to retrieve specific types of physiological data and/or result data from the data storage device, and for browsing or manually adding, deleting or modifying a marker on a set of the physiological data, and for selecting a type of physiological data entry for evaluation of a correlation with a marker;
wherein the data storage device provides automatic indexing capability, to automatically index a set of physiological data and/or result data, and wherein the data editing device is configured to display physiological data in an arrangement according to a corresponding index in response to an input request; and
a correlation evaluation device to calculate a correlation value of a type of physiological data and a marker.
The multiple physiological data collection and analysis device of the present invention may further comprise an automatic analysis device that provides a filtering interface to receive a filtering instruction and to automatically retrieve corresponding physiological data and/or result data from the data storage device, and to discover from the multiple physiological data a type of physiological data that is correlated to a specific marker.
In a preferred embodiment of the present invention, the physiological data stored in the data storage device correspond to a plurality of person and are classified into four categories: “signal-featured” physiological data, “multi-lead signal-featured” physiological data “frame-featured” physiological data and “multiple frame-featured” physiological data. Each set of physiological data is indexed with the following features:
For “signal-featured” physiological data and “multi-lead signal-featured” physiological data: file name, recording time and an identification code (ID code).
For “frame-featured” physiological data and “multiple frame-featured” physiological data: file name, recording time and an identification code (ID code).
Among these features, the file name preferably includes a personal ID code of the person from whom the physiological data set was gathered. The recording time can include a time point or a time period defined by a start time and an end time. As for the ID code, it is preferably a unique code and is preferably related to the type of physiological data included in the corresponding data set. The code length should be moderate, that is, it should not be too short to easily repeat with the ID code of another person or data set, and it should not be too long, which increases processing complexity, resources and time. In the preferred embodiments of the present invention, the ID code may comprise a hash value, especially the “Secure Hash Algorithm 256-bit” (SHA256) value, calculated according to the numerical value of the physiological data of a corresponding data set.
Mainly because of the unique data and information classification methods and specially designed indexing methods of the present invention, data in different forms, with different properties, in different storage or transmission media, and with different data volumes, data relating to different people and recorded at different times, can all be stored in a single storage device and can be retrieved, filtered, edited and otherwise utilized using a single display interface or human-machine interface, whereby possible correlations among a plurality of data set can be immediately shown or revealed. In addition, the possible correlations between various types of physiological data and the markers attached to a set of physiological phenomena can be easily discovered from the display interface or easily recognized by the invented analysis system. The present invention is useful for skilled persons to discover a type of physiological data that may be a controlling factor or key factor of a physiological phenomenon but is yet known to the world.
Other objectives and advantages of the present invention will become more apparent from the following detailed description with reference to the accompanying drawings.
Hereinafter, several preferred embodiments of the multiple physiological data collection and analysis device of the present invention will be described with reference to the drawings. It must be noted that the descriptions and illustrations of the embodiments of the present invention are only intended to present the main features and possible implementation modes of the present invention in a brief manner. The scope of the present invention should include implementations that can be derived of or deduced by the skilled persons in the industry.
Although it is not intended to limit the scope of the present invention, the inventors have found the various types of physiological data can be classified into four categories:
Mainly based on the above findings and in combination with other unique technologies, the present invention provides a useful mechanism that can gather physiological data of different types, with different features, in different formats, and stored in different media and stored them in one single database, after suitable process, for retrieving, displaying, marking, processing them in a single interface, for further machine learning, deep learning and other processing.
The multiple physiological data may be one or more than one of the various types of physiological data of EEG, EMG, ECG, EOG, blood oxygen saturation (SaO2) and pulse, Tho/Abdo Effort, Nasal-oral air flow etc. Other information that can describe the situation of a human body, organs, tissues or a part or a combination thereof can also be applied to the present invention. The physiological data storage device 156 may be a storage device of any type, with any memory capacity, or connected in any way, such as cloud drives, external hard drives, USB memory cards, static hard drives, or even mobile phones, tablets. It may also be a laptop or desktop computer, or another server computer.
Almost all the instruments used to measure the above-mentioned physiological data available in the market have already provided Internet access capabilities. Otherwise, one can connect a sensing device to the Internet via such as a smartphone or a tablet through short-distance communication protocols such as Bluetooth to transmit the sensing results. In the prior art, installing an application program in a smart phone or tablet, or other computer devices with Internet access capabilities, i.e., the mediation device 160, to receive the physiological data from a variety of physiological data sensing devices 151-155, so that the mediation device 160 can provide the physiological data to the multiple physiological data collection and analysis device 100, is already a known technology. Detailed technology thereof is thus omitted.
The multiple physiological data collection and analysis device 100 provides a data storage device 120 in connection with the data upload device 110. The data storage device 120 provides a large volume of memory space to store the physiological data uploaded by the plurality of physiological data sensing devices 151-155 and the physiological data storage device 156. The data storage device 120 also provides a memory space to store the processing result data generated by the multiple physiological data collection and analysis device 100 after processing the stored or uploaded physiological data. The configuration of the data storage device 120 is an important technical feature of the present invention and its relevant details will be explained below.
The multiple physiological data collection and analysis device 100 further comprises a data editing device 130 that is connected to the data storage device 120 and provides a human-machine interface 131 for the user to retrieve specific physiological data and/or processing results from the data storage device 120, for browsing, manually marking or modification of markers. The human-machine interface 131 may include one or more of input/output devices such as a display device, a mouse, a keyboard, a microphone, and a loudspeaker, and may also include other tools that can add, delete, and change content in a physiological data file. The human-machine interface 131 of the data editing device 130 provides a retrieval tool for users to input indices to call out one or more physiological data sets that contain corresponding indices, and to display the physiological data on the human-machine interface 131 in a predetermined form and arrangements, for the users to edit. After the user finishes editing, the processing result can be indexed and stored in the data storage device 120.
According to a preferred embodiment of the present invention, the data storage device 120 provides an automatic indexing capability, which can automatically mark and index each set of the inspection result physiological data and/or processing result physiological data. In such embodiments, the data editing device 130 is configured to retrieve a physiological data set, in response to an indexed request of a user.
According to a preferred embodiment of the present invention, the data storage device 120 of the multiple physiological data collection and analysis device 100 stores the physiological data corresponding to a plurality of person. Each set of physiological data is indexed in the following way:
In them,
[Personal ID code+type of physiological signal+device brand+device model name]. Of course, other forms of a file name may also be used in the present invention.
UuIiDdd1234-ECG-LARGAN-AT202, in which,
UuIiDdd1234=User ID
ECG=ECG signal
LARGAN=equipment manufacturer
AT202=device model
[User ID+type of physiological signal+device brand+device model name].
Of course, other forms of a file name may also be used in the present invention.
UuIiDdd1234-GLU-ABC-VP123, in which,
UuIiDdd1234=User ID
GLU=blood glucose level
ABC=equipment manufacturer
VP123=device model
In most preferred embodiments of the present invention, the hash function is chosen to calculate the ID code, mainly because the hash code is relatively short in length among all the indexing methods that are not prone to collision (different contents produce the same code value) and do not involve complicated calculations. In particular, the SHA256 code is only 256 bits long, therefore is highly suitable as a database index. In calculation, only bit reversal (XOR), shift (SHIFT), and rotation (ROT) are used; it is efficient and easy to implement. The advantages of using the SHA256 hash code as the index of a physiological data set include:
The physiological data uploaded by the data uploading device 110 are processed as described above and then saved in the data storage device 120 for later use.
As described above, the data editing device 130 of the present invention is configured to determine the relevance of different sets of physiological data, in particular, based on the index of each data set, and display the multiple physiological data that are determined to have relevance as the retrieval result.
In the foregoing steps, the correlation of two sets of data may be determined, when they have a time slot in common. For example, a plurality of sets of data whose recording time falls within a certain time period may be determined as correlated. Other methods that can determine the relevance based on the content of the data file, especially the relevance based on an element/component of the indices of a physiological data set, can also be applied to the present invention.
As for the best frame of frame-featured data, it usually refers to the data that the searcher is most likely interested. Therefore, it can also be determined based on its time feature. Other data content that can be determined as most suitable for display based on the content of the data file, especially based on the components of the indices, can also be determined as the best frame.
Specifically, the method for describing the frame-featured physiological data and the signal-featured physiological data is different. The frame-featured physiological data need to describe a value, and to define its dimension and precision (resolution). The signal-featured physiological data on the other hand adds a description of the sampling rate and the filtering method, and requires more attention on the dynamic range of changes. The multiple frame-featured physiological data and the multi-lead signal-featured physiological data are essentially frame-featured physiological data and signal-featured physiological data, respectively, provided, however, that the data included therein cannot usually be recorded and read separately. They are configured into multiple/multi-lead, mainly to facilitate simultaneous access and recording. For example, the ECG signal of 5 leads usually needs to be viewed in parallel at the same time. It is not meaningful to look at a lead alone. Dividing it into 5 independent data sets during recording would simply lead to low efficiency.
The frame-featured physiological data and the signal-featured physiological data are different in data processing and use. The frame-featured physiological data are only a point in time. Although the values of a set of data outside this time point are unknown, the values can be estimated from the values measured beforehand and afterward. For example, if there is only one white spot on the chest X-ray taken a year ago, and there is only one white spot on the chest X-ray taken today, it can be presumed that in all chest X-ray taken in the past year there should be only one white spot.
On the other hand, the signal-featured physiological data occupy a continuous section on the time axis. Only the measured values of an approximate time section or an intersection can be used as reference. For example, a patient wears an oximeter from 20:00 last night to 5:00 this morning. If his/her sleep disordered breathing index for from 22:00 to 8:00 needs be analyzed, the blood oxygen readings of the intersection between 22:00 and 5:00 can be used.
Many times, people want to find a causal relationship between a signal and a frame. For example, when a specific event (frame) occurs, people want to know if it will be accompanied by a continuous signal (signal) with specific symptoms. A good example is, experts have discovered through observation that when a sleep apnea event occurs, the heart rate will first decrease and then increase. As long as the heart rate is monitored for signs of first decreasing and then increasing, it can be used to assess whether a sleep apnea event has occurred. In this respect, the present invention can provide useful information. Through machine learning, it is possible to discover the correlation between heart rate changes and sleep apnea events. After verification, new analysis methods can be discovered.
Then, in step 530, the data editing device 130 displays the retrieved data on the human-machine interface 131 in a predetermined format. The form of display is usually images, especially graphics. However, other forms of data display, such as text, sound, animation, continuous images or discontinuous images, are also applicable.
In step 540, the data editing device 130 determines whether the user has marked or modified a manual marker. If YES, in step 550, the changes made by the user is stored in a data file that is the same as or different from the corresponding data file being displayed, and the displayed content is changed accordingly. The step returns to 540. If the judgment result of step 540 is NO, then it is determined in step 560 whether new retrieval condition are input. If YES, the step returns to 510; otherwise, it is determined in step 570 whether to end the editing. If NOT, the step returns to 540; otherwise, the editing ends in step 580. In the above steps, researchers can easily discover a possible relation between/among various types of physiological data and/or the correlation of a type of physiological data and specific physiological phenomena from the displayed information.
In terms of application, when researchers find physiological phenomena that arouse interest, they can mark manual markers on them. The manual markers can be an icon or a string of words. The data editing device 130 automatically attaches the manual markers to the physiological data file for future use.
The retrieval result in
3. Type of data: The type of the data, such as EEG, EMG, ECG, EOG, SaO2 and pulse, Tho/Abdo Effort and Nasal-oral Air Flow. Other types of physiological data, or even other categorization methods, can also be applied to the present invention.
It can be seen from
The multiple physiological data collection and analysis device 100 of the present invention may also include an automatic analysis device 140. The automatic analysis device 140 provides a filtering function and receives a filtering command from a user through the filtering interface 141, to retrieve from the data storage device 120 physiological data and/or processing result physiological data corresponding to a filtering condition included in the filter command. The filtering result data are useful for machine learning, in discovering algorithms that can be executed by a computer system, or for AI deep learning, to find out a type of physiological data that is correlated to a manual marker, i.e., a physiological phenomenon. Researchers can provide the filtering results to a machine learning program, and use approaches such as try-and-error to find out an algorithm that can be interpreted by the machine. Researchers can also provide the filtering results to an AI deep learning program, to find out a type of physiological data that is related to a manual marker.
The analysis techniques suitable for the automatic analysis device 140 of the present invention include various deep learning techniques. Existing deep learning technologies can already assist in finding from a database containing a large quantity of data specific types of physiological data that may be related to specific physiological phenomena. For example, Sun et al. proposes a methodology applicable to the present invention. See Haoqi Sun et al., Sleep staging from electrocardiography and respiration with deep learning, Sleep staging from electrocardiography and respiration with deep learning, https://pubmed.ncbi.nlm.nih.gov/31863111/). Other experts in this technical field have also proposed several technologies that can be applied in the present invention, which are all included herein for reference.
What is notable is, since the present invention has provided a simple and graphical interface that can display different types of physiological data and specific physiological phenomena on the same screen, the relevance of a type of the physiological data and certain manual markers can easily catch the attention of an observer. Researchers only need to try multiple times to retrieve different combinations of physiological data and verify their correlation with certain physiological phenomena (markers). It is possible to find a connection between certain types of physiological data and physiological phenomena that was unknown before. In other words, the human-machine interface provided by the data editing device 130 of the present invention is a tool that makes it easy for researchers to see the correlation between certain types of physiological data and certain physiological phenomena with their naked eye. With manual selection of samples, researchers may discover new algorithms for monitoring, diagnoses, treatment and/or improvements of bodily disorders.
The multiple physiological data collection and analysis device 100 of the present invention provides a correlation evaluation device 150 for calculating the correlation value of a specific type of physiological data and a specific manual marker. After the user inputs a type or a combination of types of physiological data of the filtering results in the filtering interface 141, the correlation evaluation device 150 retrieves a specific range of physiological data from the data storage device 120, and calculates a correlation value of the type of physiological data and a manual marker that was input by the user also in the filtering interface 141. The evaluation results are then displayed in a numerical or graphical form.
In step 920, the automatic analysis device 140 displays the filtering results on the filtering interface 141. In this step, the automatic analysis device 140 may provide the user with the following filtering functions:
The above filtering conditions are not in a certain order. It's acceptable to omit or add one or more filtering conditions. What is important is to find the right amount of relevant physiological data to save time in machine learning or deep learning.
In step 930, the user inputs the controlling physiological data of the filtering result into the filtering interface 141. In step 940, the correlation evaluation device 140 generates a result, which may be a presumed relevance of a controlling physiological data and a physiological phenomenon. The correlation value of the two is then evaluated. The controlling physiological data may include signal-feature and frame-featured physiological data, while the physiological phenomenon is usually a disease or a physiological abnormality. If the evaluation result is “highly correlated,” it means the finding is successful, and the result is stored in step 950. A new analysis is added or updated to the multiple physiological data collection and analysis device 100. Otherwise, the step returns to 930 or 910 for further filtering.
In the followings, specific examples are used to illustrate how researchers use the invented multiple physiological data collection and analysis device to discover and verify the correlation of a specific type of physiological data and specific physiological phenomena. In this embodiment, the readings obtained by PSG (Multiple Physiological Examination of Sleep) are used as entry, for the invented multiple physiological data collection and analysis device to perform machine learning in an attempt to discover and establish a new algorithm for sleep staging and sleep respiratory analyses.
As shown in the figure, in step 1010, various PSG detection results are input into the system in the form of Signal (signal-featured) data and Frame (frame-featured type) data for machine learning and evaluation. The signal-featured physiological data used in this example include: EEG, EMG, ECG, EOG, blood oxygen saturation (SaO2) and pulse, Tho/Abdo Effort, Nasal-oral air flow, microphone voice, body movement, leg movement etc. As for frame-featured physiological data, they include records such as manual markers for sleep staging and manual markets for respiration events. Among them, the relevant manual markers are those marked by professionals in the relevant physiological data files using the multiple physiological data collection and analysis device of the present invention.
In step 1020, one or several types of signal-featured physiological data with higher correlation values to the manual markers are found. In this step, it is preferable to use an AI-equipped computer to execute a deep learning algorithm to find the best features in the above-mentioned signal-featured physiological data relative to a specific manual marker, followed by sorting the correlation values according to a recognizability of the best features against the manual markers. In this example, it is easy for any user to identify the following possible correlations from the displayed results:
For sleep staging related manual markers, the types of physiological data are arranged in descending recognizability order as: EEG>ECG>Snout and nose airflow>Chest movements> . . . .
For respiration event related manual markers, the types of physiological data are arranged in descending recognizability order as: blood oxygen>oral and nasal air flow>ECG>chest undulation> . . . .
In more detail, taking the ECG as an example, the best feature for sleep staging is heart rate variability (time domain). The best feature for sleep breathing events is heart rate variability (frequency domain). In this way, one or several possible combinations can be presumed, that is, the correlation of specific types of physiological data, or even certain parameters thereof, and specific manual markers can be established or presumed.
In this embodiment, the recognizability may be quantized as a correlation value, which in turn can be determined by an AUC value. If the AUC value is used to represent the recognizability, the closer the value is to 1, the better. If the value is below 0.6, the correlation is considered insufficient and the type of physiological data is not selected as a controlling feature. As for the parameters selected for specific types of physiological data in verifying their correlation values, they can be selected by first referring to the suggestions mentioned in the literature. The system of the present invention can then use deep learning to verify the correlation value of the parameters and to find useful parameters not mentioned in the literature.
Next, in step 1030, a type of signal-featured physiological data for entry is selected. As mentioned above, the selection method is preferably manual selection. In this example, considering the convenience of the user's operation and the relevance to the previous steps, two types of data such as blood oxygen and ECG can be selected as entries. The main reason for choosing ECG instead of EEG in this example is that EEG has poor recognizability for sleep breathing events. At the same time, there are more electrodes for measuring EEG. The subjects cannot stick it on their own, and they tend to fall off during sleep. On the other hand, EEG is only suitable for use in situations where someone is supervised by others. In the context of home measurement, ECG is preferred. In step 1040, machine learning is performed using the features found in step 102 to train a machine learning model. In step 1050, the result of the machine learning is recorded. In a preferred embodiment, it can be recorded in the form of frame-featured physiological data.
In step 1060, the performance of the algorithm so found is evaluated. Compare various indicators of manually marked frame-featured physiological data with machine algorithm marked frame-featured physiological data:
Sample distribution: Statistical sample distribution (gender, respiratory problem degree) from the frame-featured training samples, with the results obtained as follows. It is determined that the samples used in this embodiment are representative:
Sensitivity and specificity: Compare the correlation value of the machine learning markers to the manual markers under different respiratory disorder indices (Apnea-Hypopnea Index, AHI) (AUC=accuracy, TP=true positive, FN=false negative, FP=false positive, TN=true negative). The results obtained are as follows:
In step 1070, the correlation value of the obtained analysis method is used to judge whether the found algorithm is useful. The calculation result can be expressed numerically or graphically. For example,
It should be understood that processes and techniques described herein are not inherently related to any particular apparatus and may be implemented by any suitable combination of components. Further, various types of general purpose devices may be used in accordance with the teachings described herein. The present invention has been described in relation to particular examples, which are intended in all respects to be illustrative rather than restrictive. Those skilled in the art will appreciate that many different combinations will be suitable for practicing the present invention.
Moreover, other implementations of the invention will be apparent to those skilled in the art from consideration of the specification and practice of the invention disclosed herein. Various aspects and/or components of the described embodiments may be used singly or in any combination. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the invention being indicated by the following claims.
Number | Date | Country | Kind |
---|---|---|---|
110127399 | Jul 2021 | TW | national |