Method for identifying physiological steady state and transition state of human body

Abstract

The present application provides a method for identifying a physiological steady state and a transition state of a human body, comprising: obtaining modeling features by means of physiological signals collected from a subject for at least 24 hours, and using the modeling features to construct a historical state matrix H0; constructing an individualized physiological state characterization matrix {D(Sl)}; reconstructing observed data Xobs to obtain estimated data Xest, and matching a state sequence; and calculating a residual sequence R between the observed data Xobs and the estimated data Xest, wherein if the residual sequence R satisfies a low-variance zero-mean Gaussian distribution, the physiological system is determined to be in an original steady state; if the variance of R presents a progressively increasing trend, the state is a transition state; if the mean value of R deviates from zero but is relatively stable, and the variance does not change greatly, it indicates that a physiological state that is not observed during the construction of H0 appears, and the individualized physiological state characterization matrix {D(Sl)} needs to be updated; and if R first presents a progressively increasing trend and then presents a relatively stable state after lasting for a period of time, a new steady state is determined to appear in the physiological system.

Description

A method for identifying the physiological steady state and transition state of a human body TECHNICAL FIELD

[0001] The present application relates to the field of physiological state monitoring and analysis, and in particular to a method for identifying the physiological steady state and transition state of a human body. BACKGROUND

[0002] In the fields of medicine, sports science, psychology, and biofeedback, understanding and distinguishing the physiological state of the human body is of great significance for diagnosis, treatment, and training. In recent years, the concept of "steady state" has been gradually applied in the field of physiological health. The "homeostasis" theory is an important supporting theory in physiology, which refers to the state in which the body maintains a relatively stable state through negative feedback mechanisms under normal physiological conditions. When the body experiences stress, it can adaptively adjust the target set point to adjust the body to a new homeostasis in the new environment. In this process, a "transition state" from one steady state to another new steady state is produced. Good transition can improve the body's ability to adapt to the environment and improve the level of homeostasis, otherwise it will reduce the level of homeostasis and cause damage to the body.

[0003] The concept of "steady state" of the human physiological system has been proposed for a long time, providing a theoretical basis for the identification of the physiological state of the human body. However, its abstract nature makes it difficult to achieve quantitative analysis in practical applications. In existing physiological state monitoring technologies, most methods focus on evaluating the health status of individuals through static or intermittent data collection, such as extracting time domain, frequency domain, standard template correlation, and other indicators based on physiological signals such as electrocardiogram, respiration, and activity level. Through signal processing or machine learning methods, the body state under specific conditions is analyzed and evaluated. These methods often lack a deep understanding of the continuity and dynamic changes of individual physiological states.

[0004] The patent with application number 202110245934.1 discloses a physiological state evaluation method and device, which can convert electrocardiogram data into digital integrated data for evaluation, but may not fully consider the dynamic changes of individual steady state and transition state.

[0005] The patent with application number 202310297907.8 discloses a perinatal maternal physiological state monitoring and evaluation method and system, which can monitor the physiological state of the mother in real time according to the perinatal characteristic attribute parameters, and obtain the risk assessment information of the mother through model evaluation, but may not be suitable for the wide evaluation of other physiological states.

[0006] The patent with application number 202310579449.7 discloses a vital sign data processing method and system based on intelligent wearable devices, which can evaluate the physiological state level, but may have limitations in individualized modeling and state recognition accuracy.

[0007] Disclosure of the Invention

[0008] In view of the above problems, the present application aims to provide a method for identifying the physiological steady state and transition state of a human body, which can accurately identify and quantify the steady state and transition state of an individual and adapt to the evaluation requirements under different physiological conditions.

[0009] The method for identifying the physiological steady state and transition state of a human body according to the present application comprises:

[0010] Obtaining modeling features from at least 24 hours of physiological signals collected from a subject, and constructing a historical state matrix H0 using the modeling features;

[0011] Constructing an individualized physiological state representation matrix {D (Sl)} using a time series subspace segmentation technique and external auxiliary information;

[0012] Reconstructing the observation data X obs based on a structured sparse coding framework to obtain estimated data X est , and matching the state sequence;

[0013] Calculating the residual sequence R between the observation data X obs and the estimated data X est ;

[0014] If the residual sequence R satisfies a low-variance zero-mean Gaussian distribution, it is determined that the physiological system is in the original steady state; if the variance of R shows a gradual increasing trend, the state is a transition state; if the mean of R deviates from zero but is relatively stable, and the variance does not change much, it indicates that a physiological state that was not observed when H0 was constructed appears, and the individualized physiological state representation matrix {D (Sl)} needs to be updated; if R first shows a gradual increasing trend and then shows a relatively stable state after a period of time, it is determined that the physiological system has entered a new steady state.

[0015] Preferably, the physiological signals include electrocardiogram signals, respiration signals, and blood oxygen saturation signals.

[0016] Preferably, the modeling features include basic vital sign parameters and derived parameters from the physiological signals.

[0017] Preferably, the external auxiliary information includes human posture and activity state information, and sleep staging information.

[0018] Preferably, the modeling features include heart rate, heart rate variability, respiratory rate, respiratory rate variability, pulse rate, blood oxygen saturation, cardiopulmonary coupling, body posture, and activity intensity.

[0019] Preferably, the modeling features are processed by removing noise and outliers, missing value imputation, and aggregation analysis.

[0020] Preferably, the activity state of the subject is divided into static, low-intensity activity and high-intensity activity states according to different activity intensities; and the historical state matrix H0 is divided into subspaces in different activity states according to the activity state information of the subject.

[0021] Preferably, the sleep staging information is obtained by sleep staging detection, including: light sleep, deep sleep and REM state; and the historical state matrix H0 is divided into subspaces in different sleep states according to the sleep staging information of the subject.

[0022] Preferably, the reconstruction of the observation data X obs is achieved by solving an optimization problem.

[0023] The method for distinguishing between the transition state and the steady state of the human body in the present application preprocesses physiological signals such as electrocardiogram, respiration, body position / body movement collected by a wearable device, then extracts features, and individualizes modeling by the method for estimating the multiple states of the human body, distinguishes between the transition state and the steady state of the human body by matrix construction, data reconstruction and residual analysis, and provides technical support for individualized physiological homeostasis monitoring, and is expected to further improve the application value of physiological data of a wearable device, and provide quantitative analysis support technical means for individualized disease management, rehabilitation training and performance improvement.

[0024] Brief description of the drawings

[0025] Fig. 1 is a flowchart of the method for distinguishing between the transition state and the steady state of the human body in the present application;

[0026] Fig. 2 is a physiological sequence state division diagram based on external auxiliary information;

[0027] Fig. 3 is a flowchart of the human body transition state and steady state recognition technology;

[0028] Fig. 4 is a residual change diagram of a high-altitude hypoxia experiment.

[0029] Best mode for implementing the present application

[0030] The implementation environment of the present application includes but is not limited to medical institutions, sports training centers, family or personal health monitoring and the like. The required devices include wearable devices, non-invasive physiological signal sensors, data processing and analysis software.

[0031] Here, it should be noted that the contents of each module of the present application are the corresponding functional modules realized by the method for distinguishing between the steady state and the transition state of the human body in the present application when running on a computing device (such as a smart phone, a tablet, a computer or a server, etc.).

[0032] The method for distinguishing homeostasis and transition state of human body is described in detail below with reference to the accompanying drawings.

[0033] 1. Collecting physiological signals

[0034] Select appropriate wearable devices or non-invasive sensors such as electrocardiogram (ECG) sensors, respiratory bands, oximeters, etc. to continuously collect physiological signals for at least 24 hours, and remove unusable data segments through a signal quality evaluation algorithm. The collection and quality evaluation of physiological data are shown in FIG. 1.

[0035] 2. Modeling feature calculation and screening

[0036] Extract a set of features from the original physiological signals every t minutes, including heart rate, heart rate variability, respiratory rate, respiratory rate variability, pulse rate, oxygen saturation, cardiopulmonary coupling, body posture, activity intensity, etc. as modeling feature parameters. Further process the modeling feature parameters, including removing noise and outliers, missing value interpolation, aggregation analysis, etc. to screen out usable parameters and form a historical state matrix H0 that can be used for modeling.

[0037] 3. State division of H0 based on external auxiliary information

[0038] State division based on body position / body movement information is shown in FIG. 2a. Quantify activity intensity through three-axis accelerometer signals, divide into static, low-intensity activity and high-intensity activity states according to different activity intensities, and divide H0 into subspaces in different activity states.

[0039] State division based on sleep information is shown in FIG. 2b. Perform sleep staging detection by integrating various physiological signals, divide into light sleep, deep sleep and REM state, and divide H0 into subspaces in different sleep states.

[0040] In addition, according to the data situation, H0 can also be segmented according to the time series subspace separation technique, and after normalization, a plurality of subspaces {D (1) ,D (2) ,…D (m)} are obtained, and a state matrix {D (Sl)} is generated. Here, {D (Sl)} = A(H0), A represents a certain algorithm or rule, which is used to map H0 into a plurality of state subspaces.

[0041] 4. Reconstructing new observation data X obs :

[0042] For new observation data X obs , select X (Sl) from {D obs} according to the size of the Gaussian kernel.The top-scored samples constitute a matching matrix D obs The matching state sequence S obs corresponding to the matching matrix D is simultaneously generated

[0043] The reconstruction result is obtained by solving the following optimization problem:

[0044] where C is the encoding coefficient matrix, λ > 0 is a compromise parameter, D = {D (1) , D (2) , …, D (m)}, or

[0045] where δ > 0 is a certain allowed interpretation error level, or

[0046] where α > 0 is a certain constraint on the interpretation rationality, and r(C) is a structured sparse regularization term corresponding to the structure in D. By solving the optimization problem, the observed data is reconstructed to form the estimated data X est .

[0047] 5. Calculate the residual:

[0048] Calculate the residual vector sequence R = X obs - X est .

[0049] 6. Transient state and steady state identification

[0050] The state identification flowchart of the human transient state and steady state is shown in Fig. 4. If the residual sequence R satisfies the low-variance zero-mean Gaussian distribution, and the matching state sequence S obs has stationarity and consistency, it indicates that the observed data X obs of the system in the steady state can be effectively explained by the individualized physiological state representation matrix {D (Sl)}, and it can be judged that the physiological system is at the original steady state level; otherwise, if

[0051] (a) the mean of the residual sequence R is still close to zero, but the variance is significantly increased, it is a transient state;

[0052] (b) the mean of the residual sequence R deviates from zero significantly, but the variance has no obvious change, which means that a physiological state that is not observed in H0 appears, and at this time the individualized physiological state representation matrix {D (Sl)} needs to be updated.

[0053] Examples

[0054] A specific example is provided below to demonstrate the application of the method of the present application:

[0055] Embodiment environment:

[0056] Select a volunteer to conduct a simulated high-altitude oxygen cabin simulation experiment, including about 20 hours of baseline physiological data on the plain and about 2 hours of simulated high-altitude data, and continuously collect the physiological signals of the volunteer using a wearable device.

[0057] Data processing flow:

[0058] The first 2 hours of baseline data on the plain are used as initial data, preprocessed and feature extracted with a 1-minute window, the features include heart rate, respiratory rate and activity-related parameters, and H0 is constructed. H0 is segmented into D by sleep staging and activity level. obs .

[0059] The remaining baseline data on the plain is used as X obs , and any single observation vector in X obs is matched with the corresponding D obs according to its sleep and activity level rating, and the optimal 10 historical vectors are reconstructed to obtain the reconstructed value X est , and the residual R is calculated.

[0060] Result analysis:

[0061] After entering the simulated high-altitude environment, as shown in FIG. 4, it is found through residual analysis that the R variance increases significantly, and it is determined to be a transition state.

[0062] After a period of time, the R variance decreases and stabilizes, and after updating {D (Sl)}, it is determined to be a new steady state.

[0063] Implementation effect:

[0064] Through the above embodiment, the method of the present application can effectively distinguish the steady state and the transition state of the human body, and provides a new technical means for individualized physiological state monitoring. Industrial applicability

[0065] The method of the present application can distinguish the steady state and the transition state of the human body through individualized modeling and analysis of continuous physiological data. The individualized modeling method of the present application solves the limitations of traditional methods in physiological state recognition, improves the accuracy and individual adaptability of state recognition; the construction of the individualized physiological state representation matrix proposed by the present application can solve the state aliasing problem and improve the accuracy of state recognition; the reconstruction and residual analysis method of new observation data used by the present application can realize the sensitivity to small state changes and the fitting ability to complex human physiological system states through data driving; the dynamic updating mechanism of state recognition proposed by the present application can dynamically update the individualized physiological state representation matrix according to the residual analysis result to adapt to the changes of human physiological state. The application results of the present application show that the method of the present application can effectively distinguish the changes of human state, and the test results are good, which is suitable for a wide range of application scenarios, including but not limited to medical health monitoring, sports training, psychology research and biofeedback fields.

[0066] Unless otherwise defined, all technical and / or scientific terms used within the present application have the same meaning as commonly understood by one of ordinary skill in the art to which the present application pertains. The materials, methods, and examples mentioned in the present application are merely illustrative, rather than limiting.

[0067] Although the present application has been described in conjunction with the specific embodiments thereof, those skilled in the art will be able to make appropriate substitutions, modifications and changes under the spirit of the present application, and such substitutions, modifications and changes still fall within the scope of protection of the present application.

Claims

1.A method for identifying a human body in a physiological steady state and a transition state, comprising: obtaining modeling features from physiological signals collected from a subject for at least 24 hours, and constructing a historical state matrix H0 using the modeling features. By using the time series subspace segmentation technique and external auxiliary information, an individualized physiological state representation matrix {D (Sl)} is constructed. On the observed data X obs perform reconstruction based on structured sparse coding framework to get the estimated data X est and match the state sequence; The computed observation data X obs The residual series R between the estimated data X est and the computed observation data X If the residual sequence R satisfies the low-variance zero-mean Gaussian distribution, it is judged that the physiological system is in the original steady state; if the variance of R shows an increasing trend, it is in a transition state; if the mean of R deviates from zero but is relatively stable, and the variance changes little, it indicates that a physiological state not observed in the construction of H0 appears, and the individualized physiological state representation matrix {D (Sl)} needs to be updated; if R first shows an increasing trend, and then shows a relatively stable state after a period of time, it is judged that the physiological system has a new steady state. 2.The method of claim 1, wherein: the physiological signals include electrocardiogram signals, respiration signals, and blood oxygen saturation signals. 3.The method of claim 2, wherein: the modeling features include basic vital sign parameters and derived parameters from the physiological signals. 4.The method of claim 1, wherein: the external auxiliary information includes posture and activity state information and sleep staging information. 5.The method of claim 2, wherein: the modeling features include heart rate, heart rate variability, respiration rate, respiration rate variability, pulse rate, blood oxygen saturation, cardiopulmonary coupling, body posture, and activity intensity. 6.The method of claim 5, wherein: the modeling features are processed by noise and outlier removal, missing value imputation, and aggregation analysis. 7.The method of claim 4, wherein: activity states of the subject are divided into static, low-intensity activity, and high-intensity activity states according to different activity intensities, and the historical state matrix H0 is divided into subspaces in different activity states according to the activity state information of the subject. 8.The method of claim 4, wherein: sleep staging information is obtained by sleep staging detection, including light sleep, deep sleep, and REM state, and the historical state matrix H0 is divided into subspaces in different sleep states according to the sleep staging information of the subject. 9.The method of claim 1, wherein: The reconstruction of the observed data X obs is achieved by solving an optimization problem.

Images