A community medical-oriented remote central monitoring system

Abstract

The present application belongs to the technical field of medical care information science, and discloses a community medical remote central monitoring system, which comprises a data acquisition and integration module, a physiological state analysis engine and a monitoring information presentation module.The system can construct a state-dependent baseline model cluster based on the activity state information of a user, dynamically select a corresponding sub-model as a judgment benchmark according to the real-time activity state of the user in a monitoring period, and then generate a continuously changing physiological stability index by comparing the deviation of the multi-dimensional statistical characteristics of real-time physiological data from the benchmark.The present application establishes an individualized physiological steady state evaluation benchmark that is self-adapted to activity contexts, improves the ability of the monitoring system to continuously quantitatively evaluate the overall stability of the human body as a complex system, and realizes the transition from passive detection of discrete clinical events to state insight into the risk evolution process.

Description

TECHNICAL FIELD

[0001] The present application relates to a community-oriented remote central monitoring system, belonging to the technical field of medical care informatics. BACKGROUND

[0002] The current community-oriented remote monitoring system is generally based on a mechanism that sets a threshold for a single physiological parameter of a user to trigger an alarm. This mechanism is directly effective in capturing acute and discrete physiological events and is the technical basis that has been widely used. However, when this technical approach is applied to large-scale uncontrolled home monitoring, an inherent constraint in its design principle, i.e., its role as a disruptive detection of physiological homeostasis, leads to structural limitations in its information processing capacity under new application conditions. The reason for this is that the deterioration of the human physiological state is usually characterized by a gradual disorder in the coordination between multiple parameters, and the overall stability of the system continues to decline within the threshold range. The existing technical approach has inherent information gaps for this process-oriented risk evolution.

[0003] To solve the above problems, the technical personnel have attempted to lower the sensitivity of the alarm threshold in order to capture earlier signals, but in the home environment full of daily activity interference, this has caused a sharp increase in false alarms, resulting in the problem of alarm fatigue that is difficult to handle clinically. Another attempt is to collect and transmit more comprehensive raw waveform data to the central end, but this puts forward too high economic requirements for the power consumption, computing power and communication bandwidth of the user-side device, making it not universal. The limitations of these attempts show that the core of the problem is not the accuracy of the sensor or the amount of data, but the existing information processing method itself, which can only process the absolute value of the physiological parameter, but lacks a mechanism for quantitatively evaluating the relative stability of the overall system state represented by the multidimensional physiological data. For example, Chinese invention patent CN116980438A discloses a remote medical monitoring system and method, which focuses on solving the data transmission security problem in remote monitoring by setting an isolation architecture of an external network server, a network gate, and an internal network server. However, the core of this scheme is still to transfer the collected physiological parameters to the central workstation for display, and its technical focus is on building a secure data pipeline, which does not fundamentally change the passive interpretation of raw data by medical personnel, and does not solve the inherent information processing limitations of the traditional threshold alarm mechanism, and cannot realize intelligent evaluation of the risk of physiological state.

[0004] The prior art has the following deficiencies: 1. The information model is not matched with the progressive evolution law of the human physiological state in the home monitoring scene, and there is a lack of effective monitoring means for the transition from a steady state to an unstable state; 2. The working mode is a post-event detection, and it is impossible to provide risk assessment information for clinical intervention; 3. It is difficult to be compatible between the two technical requirements of pursuing high early warning sensitivity and maintaining low false alarm rate. Therefore, how to establish an information processing and presentation architecture, so that it is no longer dependent on passive detection of isolated parameter thresholds, but can continuously quantitatively evaluate the stability of the coordinated dynamics of multi-dimensional physiological data based on the individualized physiological baseline model of the user, and convert the evaluation results into information that can be directly used for risk ranking of multiple monitoring objects, becomes a technical problem to be solved by the present application. SUMMARY

[0005] The present application provides a remote central monitoring system for community medical treatment, which mainly aims to solve the problem that the information processing mode of the existing remote monitoring technology is not matched with the progressive evolution law of the human physiological state, and the overall stability of the system cannot be evaluated.

[0006] To achieve the above-mentioned purpose, the present application provides a remote central monitoring system for community medical treatment, comprising:

[0007] a data acquisition and integration module configured to receive activity state information and multi-dimensional physiological data from an activity state sensor and at least one physiological sensor of a user, respectively;

[0008] a physiological state analysis engine configured to perform the following rules: identifying a plurality of typical activity states of the user based on the activity state information; and for each typical activity state, calculating a set of statistical characteristics based on the multi-dimensional physiological data collected in the corresponding typical activity state, to construct a set of individualized physiological baseline sub-models corresponding to the plurality of typical activity states one by one, the set of statistical characteristics comprising the mean value of each dimension of the multi-dimensional physiological data, the standard deviation of each dimension, and the correlation coefficient between any two dimensions, and a state-dependent baseline model cluster composed of the set of individualized physiological baseline sub-models; during the monitoring period, dynamically selecting an individualized physiological baseline sub-model corresponding to the current activity state from the state-dependent baseline model cluster as a judgment reference according to the real-time monitored current activity state of the user; continuously collecting real-time multi-dimensional physiological data in a sliding time window, calculating a set of real-time statistical characteristics identical in dimension and type with the set of statistical characteristics of the individualized physiological baseline sub-model; and calculating the deviation degree between the real-time state vector defined by the set of real-time statistical characteristics and the baseline vector defined by the set of statistical characteristics of the judgment reference by a preset distance formula, and converting the deviation degree into a continuously changing physiological stability index;

[0009] a monitoring information presentation module configured to receive physiological stability indices from a plurality of users, and to continuously and automatically dynamically re-rank positions of the plurality of users on a monitoring interface according to values of the physiological stability indices.

[0010] Preferably, the physiological state analysis engine is further configured to perform the following rules: continuously monitor the physiological stability index it generates itself; and when it identifies that the physiological stability index continuously exceeds a preset stability threshold for a preset duration, automatically trigger a baseline model updating process, and calculate a candidate baseline model using the multidimensional physiological data collected in the corresponding activity state during the duration when the physiological stability index continuously exceeds the stability threshold; and fuse the candidate baseline model with the corresponding individualized physiological baseline sub-model in the state-dependent baseline model cluster through a weighted average formula to generate an updated individualized physiological baseline sub-model.

[0011] Preferably, the physiological state analysis engine is further configured to perform the following rules: after generating the physiological stability index, determine a deviation vector from a baseline vector to a real-time state vector; and directionally compare the deviation vector with a preset prototype library containing a plurality of prototype vectors representing directions of underlying physiological disturbances to identify a prototype vector with the highest directional similarity to the deviation vector; wherein the directional similarity is calculated by the following formula: wherein, is the deviation vector, is any prototype vector in the prototype library, • is the vector dot product operation, is the Euclidean norm of the vector; and based on the identified prototype vector, attach a mode label representing the deviation mode to the physiological stability index.

[0012] Preferably, the data acquisition and integration module is further configured to obtain a signal quality index representing the quality of the output signal from at least one physiological sensor; and the physiological state analysis engine, when generating the physiological stability index, is configured to perform the following rules: within a sliding time window, discard multidimensional physiological data points with a signal quality index lower than a preset quality threshold, and do not use them to calculate the real-time set of statistical features; and calculate the average value of the signal quality indices of all data points in the sliding time window that are not discarded as a credibility factor; and multiply the physiological stability index by the credibility factor to generate a credibility-weighted physiological stability index.

[0013] Preferably, the stability threshold is set to 0.9; and the weighted average formula is: updated individualized physiological baseline sub-model = 95% x old individualized physiological baseline sub-model + 5% x candidate baseline model.

[0014] Preferably, the physiological state analysis engine uses Mahalanobis distance formula to calculate the deviation degree.

[0015] Preferably, the monitoring information presentation module is configured to perform the continuous automatic dynamic reordering, specifically, to display the users with lower physiological stability index values at a higher priority position on the monitoring interface.

[0016] Preferably, the physiological state analysis engine is deployed in a user-side edge computing device; the edge computing device is configured to send a data stream mainly carrying the physiological stability index to the monitoring information presentation module deployed in a cloud server.

[0017] Preferably, the physiological state analysis engine is configured to identify multiple typical activity states of the user based on the activity state information, specifically, to process the three-axis acceleration data stream output by the activity state sensor using an unsupervised K-means clustering algorithm to automatically cluster the activity data of the user into three typical activity states of rest, light activity and moderate activity.

[0018] Preferably, the monitoring information presentation module is further configured to display the mode label corresponding to the physiological stability index value of each user on the monitoring interface, and to retrieve and present the physiological stability index historical change curve and the original multi-dimensional physiological data curve of the user when the user selects any specific user.

[0019] Compared with the prior art, the present application has the following advantages:

[0020] 1. By constructing an individualized physiological baseline model containing mean standard deviation and correlation coefficient between different physiological data based on the multi-dimensional physiological data of the user in a specific activity state in the user-side edge computing device, continuously collecting real-time physiological data in a sliding time window and calculating the dynamic characteristics thereof, and finally generating a quantitative physiological stability index by comparing the dynamic characteristics with the baseline model, the index is a continuously changing data stream, and its core role is not to trigger an alarm, but to serve as an objective basis for dynamically ranking multiple users on the remote monitoring information presentation module. The combination of these technical features changes the focus of monitoring work from passive response to a single discrete physiological parameter threshold event to continuous and proactive assessment of the overall stability of the monitoring object as a complex system, and the attention of medical staff is thus directed to a small number of individuals with the most unstable physiological state, avoiding a large number of invalid alarm information due to isolated parameter fluctuations in the traditional way.

[0021] 2. By utilizing information collected by activity state sensors, multiple typical activity states of the user are identified within the initial calibration period. A corresponding individualized physiological baseline sub-model is constructed for each typical activity state, thus forming a state-dependent baseline model cluster. In subsequent monitoring periods, the system dynamically selects the corresponding sub-model from the baseline model cluster based on the user's current activity state monitored in real time, using it as a benchmark for calculating the physiological stability index. This dynamic partitioning and instantaneous recall mechanism of the baseline based on activity state is tightly coupled with the core physiological stability index generation mechanism, enabling the system to effectively distinguish between healthy active physiological changes and real pathological dynamic abnormalities. This allows the system to not only perform accurate stability assessments in the user's resting state but also to provide risk insight in non-resting states such as the user's daily activities, overcoming the inherent robustness limitations of a single static baseline model in real and ever-changing life scenarios.

[0022] 3. By continuously monitoring the physiological stability index stream generated by the physiological state analysis engine, when the index is identified as being in a high stability state for a preset period of time, a baseline model update process is automatically triggered. The original individualized physiological baseline model is then updated by weighted fusion using multidimensional physiological data collected during this high stability period. This process uses the system's output signal to calibrate its own reference benchmark, forming a self-consistent logical closed loop for model maintenance. This enables the individualized physiological baseline model, which serves as the cornerstone of the entire system's evaluation, to autonomously and progressively track the slow evolution of the user's true physiological homeostasis due to improved physical function or the progression of chronic diseases. This ensures the accuracy and reliability of the system during long-term continuous monitoring and avoids false alarms or missed alarms caused by outdated models.

[0023] 4. After calculating the quantified physiological stability index, this invention does not discard intermediate information from the calculation process. Instead, it further determines the deviation direction of the current physiological state from the individualized physiological baseline model and compares this deviation direction with a pre-set prototype library containing multiple basic physiological disorder direction prototype vectors. Finally, based on the comparison results, a pattern label representing its deviation pattern is added to the output physiological stability index. This reuse of deviation direction information, combined with the quantitative assessment function of the physiological stability index, makes the system output to medical staff not just a uniform numerical value indicating the degree of risk, but a collaborative information package that simultaneously contains directional hints of the inherent risk pattern. While guiding medical staff to focus on high-risk individuals, it also provides objective data pattern classification guidance for their subsequent focused analysis to explore the root causes of risk, elevating a simple data monitoring system into an information support system that can assist clinical decision-making. Attached Figure Description

[0024] Fig. 1 Flow chart for physiological stability index generation and dynamic risk assessment process of the present application;

[0025] Fig. 2 Multi-parameter trend chart for progressive risk evolution process of the present application;

[0026] Fig. 3 System edge intelligence analysis and cloud presentation architecture diagram of the present application. DETAILED DESCRIPTION

[0027] In order to make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described in detail below, but those skilled in the art can understand that the following examples and features in the examples can be combined with each other without conflict, and the specific examples described herein are only used to explain the present application, and are not used to limit the protection scope of the present application.

[0028] The remote central monitoring system for community medical treatment provided by the present application has a system architecture composed of a data acquisition and integration module deployed on the user side, a physiological state analysis engine, and a monitoring information presentation module deployed on a cloud server. The data acquisition and integration module is used to obtain user data, the physiological state analysis engine performs information processing and data conversion, and the monitoring information presentation module displays the processed information in order. The three work together to continuously quantitatively evaluate the physiological state stability of the user. In a specific embodiment, in order to deal with the interference of physiological parameter fluctuations caused by user daily activities on the monitoring results, the physiological state analysis engine of the present system is configured to first perform an initialized individualized baseline model construction process. The input data of the process is provided by the data acquisition and integration module. The module obtains the activity state information of the user through a three-axis acceleration sensor built into the edge computing device on the user side, and simultaneously receives multi-dimensional physiological data such as heart rate, blood oxygen saturation, and blood pressure through a physiological sensor of the user, such as a wrist-type oximeter or an electronic sphygmomanometer, through a wireless communication protocol such as Bluetooth. In order to ensure effective identification of the activity state, the sampling frequency of the three-axis acceleration sensor is set to 50 Hz. Within a preset calibration period of the initial operation of the system, for example, 72 hours, the physiological state analysis engine applies a K-means clustering algorithm to process the collected three-axis acceleration data stream. According to the distribution of data points in three-dimensional space, the activity state of the user is clustered into several typical activity states. For example, when the number of clusters is set to 5, the activity state of the user is clustered into five typical activity states, such as sitting, standing, walking, running, and sleeping. The physiological state analysis engine then applies a linear regression algorithm to the physiological data collected in each typical activity state to construct an individualized baseline model for each typical activity state. The baseline model is used to represent the physiological state of the user in the corresponding activity state, and the physiological state analysis engine uses the baseline model to evaluate the physiological state of the user in the corresponding activity state. In the process of continuous monitoring, the physiological state analysis engine continuously updates the baseline model according to the physiological data collected in the corresponding activity state, so as to ensure the accuracy of the evaluation results. When set to 3, the activity state can be objectively divided into resting, light activity and moderate activity; then, for the three identified typical activity states, the engine will screen the multi-dimensional physiological data collected synchronously in the corresponding state, and calculate an independent statistical feature set for each activity state to construct a corresponding individual physiological baseline sub-model; each statistical feature set includes: the mean value of each dimension of the multi-dimensional physiological data, which represents the central tendency of the parameter in the activity state; the standard deviation of each dimension, which represents the fluctuation range of the parameter; and the correlation coefficient between any two dimensions, which represents the correlation between different physiological parameters; in this way, the three individual physiological baseline sub-models corresponding to resting, light activity and moderate activity respectively form a state-dependent baseline model cluster, which provides a dynamically adjusted reference benchmark for subsequent real-time state assessment.

[0029] After entering the monitoring period, in order to continuously quantify the user's physiological stability, the physiological state analysis engine continuously collects real-time multi-dimensional physiological data with a sliding time window of 5 minutes, and dynamically selects an individual physiological baseline sub-model corresponding to the current activity state of the user from the state-dependent baseline model cluster according to the three-axis acceleration sensor, as a judgment benchmark; the engine uses the same calculation method as when constructing the baseline model to process the data in the current sliding time window, obtaining a real-time statistical feature set identical in dimension and type to the statistical feature set of the selected individual physiological baseline sub-model; the system defines the vector of the real-time statistical feature set as a real-time state vector, and the vector of the statistical feature set of the judgment benchmark as a baseline vector, and calculates the deviation of the two vectors in the multi-dimensional feature space through the preset Mahalanobis distance formula; after normalization, the deviation value is converted into a continuous physiological stability index between 0 and 1, where an index value close to 1 indicates a high degree of agreement between the current state and the individual baseline, and an index value close to 0 indicates a large deviation, i.e. the system stability decreases; considering the uncertainty of sensor signal quality in the home environment, the data acquisition and integration module is also configured to obtain a signal quality index (SQI) representing the output signal quality of the physiological sensor while receiving the physiological parameter values; before the physiological state analysis engine calculates the real-time statistical feature set, it will first perform data screening to discard the multi-dimensional physiological data points with a signal quality index below the preset quality threshold of 0.5 in the sliding time window; after calculating the deviation, the engine will also calculate the average value of the signal quality index of all non-discarded data points in the window as a reliability factor, and multiply the previously generated physiological stability index by the reliability factor to generate a reliability-weighted physiological stability index, which takes into account the reliability of the input data source in the final evaluation result.

[0030] In addition, in order to enable the individualized physiological baseline model to adapt to the long-term changes in the user's physiological state, the physiological state analysis engine also contains a model self-adaptive evolution mechanism; the engine continuously monitors the physiological stability index stream output by itself, and when it identifies that the index has been greater than the stability threshold of 0.9 for a preset length of time, for example, continuously for 48 hours, the system determines that the user has entered a new stable state, and automatically triggers a baseline model update; this update process uses the multi-dimensional physiological data collected during this high stability period to calculate a candidate baseline model, and through a weighted average formula, fuses the candidate baseline model with the corresponding old individualized physiological baseline sub-model in the state-dependent baseline model cluster to generate an updated individualized physiological baseline sub-model; the weighted average formula can be set as: updated individualized physiological baseline sub-model = 95% x old individualized physiological baseline sub-model + 5% x candidate baseline model, this kind of smooth updating method enables the baseline model to track the real steady-state evolution of the user; in order to provide prompt information about the intrinsic patterns of risk states to medical personnel, the physiological state analysis engine also performs a deviation pattern characterization step after generating the physiological stability index; this step uses intermediate information in the calculation process, that is, the deviation vector from the baseline vector to the real-time state vector; the engine performs directional comparison of this deviation vector with a prototype library pre-stored in the device, which contains a plurality of basic physiological disorder direction prototype vectors, to identify the prototype vector with the highest directional similarity to the current deviation vector; the calculation of directional similarity is carried out through the cosine similarity formula, which is: wherein, is the directional similarity; is the deviation vector; is any prototype vector in the prototype library; is the vector dot product operation; is the Euclidean norm of the vector; the engine attaches a pattern label representing the deviation pattern to the currently output physiological stability index according to the prototype vector with the highest similarity identified, for example, (psychogenic stress pattern).

[0031] The establishment of the prototype library follows a standardized data-driven procedure, which takes as input a multi-dimensional physiological and activity state history database with time-stamped and event-labeled for multiple users during the period of known clinical events (e.g. early infection or acute stress) and their preceding stable periods. The specific steps are as follows: first, for each annotated clinical event in the database, extract the multi-dimensional physiological data corresponding to the activity state within 24 hours before the event, and calculate an individualized physiological baseline sub-model representing the user's stable state. Then calculate the real-time state vector of the user during the event period, and determine a deviation vector from the baseline vector to the real-time state vector. The deviation vector is then unitized to retain only its directional information. Next, unitized deviation vectors corresponding to the same clinical event label are collected from all users into a vector set, and a density clustering algorithm such as DBSCAN is applied to process the vector set, grouping highly clustered vectors in direction into a class, and calculating the centroid vector of each class. Finally, these centroid vectors, which have been reviewed and labeled by clinical experts, are collectively solidified as a set of prototype vectors in the prototype library. Finally, the physiological state analysis engine deployed in the user-side edge computing device processes the data stream generated after processing, which takes the physiological stability index as the main load and may be attached with the mode label, and sends it to the monitoring information presentation module deployed in the cloud server. The module receives data streams from multiple users and continuously automatically dynamically reorders all users on a monitoring interface according to the value of the physiological stability index. The specific sorting logic is to display the user with a lower physiological stability index value in a higher priority position on the monitoring interface, while displaying the corresponding mode label. When a healthcare worker selects a specific user, the system can retrieve and present the user's physiological stability index history curve and original multi-dimensional physiological data curve.

[0032] In an application example of the community home monitoring, a 75-year-old male user with a history of chronic heart failure and type 2 diabetes is equipped with the remote central monitoring system of the present application; during the initial 72-hour calibration period, the data acquisition and integration module of the system obtains the activity state information and multi-dimensional physiological data of the user in daily life, and the physiological state analysis engine applies the K-means clustering algorithm to identify the activity state of the user as three typical states of rest, light activity and moderate activity, and constructs individualized physiological baseline sub-models for each state, including the mean heart rate, heart rate standard deviation, mean blood oxygen saturation and other statistical characteristics in the corresponding state, which together constitute a state-dependent baseline model cluster specific to the user; in the subsequent continuous monitoring period, the user has a mild upper respiratory tract infection and does not strictly follow the doctor's advice to take medicine, and his physiological state has slowly deteriorated over the next week; during this period, most of the user's single heart rate and blood oxygen saturation readings do not exceed the conventional alarm threshold, for example, his resting heart rate is still mostly within the range of 60 to 100 beats per minute; however, the physiological state analysis engine deployed on the user side continuously compares the real-time multi-dimensional physiological data collected with the sub-model corresponding to the current activity state in the state-dependent baseline model cluster within a 5-minute sliding time window; from the third day of the monitoring period, the real-time statistical feature set calculated by the engine starts to show a persistent deviation from the baseline vector, specifically, even if the user is in a resting state, the standard deviation of his heart rate starts to decrease compared to the corresponding value in the baseline model, at the same time, the correlation coefficient between his heart rate and blood oxygen saturation changes from stable positive correlation to unrelated; these parameters representing changes in internal coordination of the system work together to cause the Mahalanobis distance between the real-time state vector and the baseline vector to continuously increase, and then the physiological stability index generated by the system slowly and continuously decreases from the initial average value of 0.95, and by the fifth day, its daytime average has dropped to 0.70.

[0033] The monitoring information presentation module in the cloud, the monitoring interface of the medical staff has not appeared any discrete alarm event during this period; but due to the continuous decrease of the user's physiological stability index value, the built-in continuous automatic dynamic reordering logic of the system makes the user's position in the monitoring list automatically move from the middle and lower area to the high priority position day by day; by the morning of the fifth day, the user has been in the second place of the risk ranking list, and the on-duty nurse notices it when checking regularly, clicks to view the detailed information, and the interface presents not a single point overrun of a parameter, but a physiological stability index historical change curve that gently but definitely decreases in the past 48 hours; this information shows that the user's overall physiological state is in the process of evolving towards instability, and the nurse initiates remote video inquiry accordingly, and learns about the user's specific situation in the inquiry, and immediately arranges a home visit and coordinates the doctor to adjust the drug regimen, thereby pre-intervening on a potential acute decompensation event; as a further explanation of the foregoing scenario, when the on-duty nurse views the user's physiological stability index historical change curve on the monitoring information presentation module, the system also adds a mode label (cardiac stress mode) next to the key turning point of the curve and the current low index value according to the real-time calculation and pattern matching result of the physiological state analysis engine on the deviation vector; the appearance of this label is due to the system recognizing that the user's current physiological state deviation vector is highly similar in direction to the prototype vector in the prototype library pointing to the dimensions of heart rate mean increase and heart rate standard deviation decrease; this information enables the nurse to focus on the user's heart function related problems directly when initiating remote inquiry, rather than conducting open investigation, thereby improving the efficiency of subsequent clinical communication and judgment.

[0034] Example 2: To verify the performance of the present technical solution in recognizing gradual physiological state changes and coping with daily activity disturbances, a comparative test based on physiological signal simulation was designed and implemented. The test used a computational simulation system capable of generating multi-parameter physiological data streams, and the output data matched the input of the data acquisition and integration module in the specific embodiment in terms of sampling rate and signal characteristics. The test set up a test group using the complete technical solution of the present invention, and two control groups. Control group 1 used a technical method based only on a single-parameter static threshold (set at: resting heart rate greater than 120 beats / min or blood oxygen saturation less than 90%), and control group 2 used a technical method with only an individualized physiological baseline model based on a single resting state, i.e. lacking a state-dependent baseline model cluster. The first stage of the test aimed to verify the recognition ability of gradual state changes. The simulation system was configured to generate a continuous physiological data stream for 72 hours, simulating a user continuously in a resting state, with the stability of the physiological system slowly declining due to internal state changes. During this process, the absolute values of heart rate and blood oxygen saturation were always controlled within the static threshold range of control group 1, but the internal statistical characteristics, such as heart rate variability and the correlation between multiple parameters, continued to deviate from the initial healthy steady state over time. The test group and the two control groups processed the data stream respectively, and recorded key output indicators at specific time points. The results are shown in Table 1.

[0035] Table 1: Test data comparison table for gradual state change scenarios.

[0036]

[0037] From the data of Table 1, it can be seen that during the simulation of 72 hours, the output of the control group 1 is always normal state, and the state evolution cannot be recognized because the single physiological parameter value does not reach the preset threshold value; the physiological stability indexes of the systems of the control group 2 and the test group present a clear downward trend with time, and the indexes decrease to 0.53 at the 72nd hour, indicating that the system stability has decreased, which shows that the information processing mode based on the individual physiological baseline model for deviation quantization can recognize the threshold-in risk evolution which cannot be found by the conventional threshold mode; the second stage of the test aims to verify the performance of the system in response to the interference of daily activities, and the simulation system is configured to generate a physiological data stream with a length of 1 hour, which simulates a healthy steady-state user whose activity state switches between rest and light activity. During the light activity, the simulated heart rate of the user physiologically increases from 75 beats per minute in rest to 110 beats per minute, which does not exceed the alarm threshold of the control group 1; the systems of the test group and the control group 2 process the data stream respectively, and the physiological stability index output results in different activity states are shown in Table 2.

[0038] Table 2: Comparison table of test data in the activity state switching scenario.

[0039]

[0040] From the data of Table 2, it can be seen that when the simulated activity state switches to light activity, the system of the control group 2 outputs a physiological stability index as low as 0.45 due to the physiological heart rate increase caused by activity being misjudged as a serious deviation from the resting steady state, resulting in a false risk prompt; the physiological state analysis engine of the system of the test group automatically selects the individual physiological baseline sub-model corresponding to light activity from the state-dependent baseline model cluster as the judgment reference according to the activity state information. Since the heart rate of 110 beats per minute is within the normal range in the light activity baseline model, the calculated physiological stability index of the test group still maintains a high level of 0.92, accurately reflecting the stability of the user in this activity state; the comparison results show that the state-dependent baseline model cluster can avoid misjudgment caused by user daily activities; the results of this test confirm that the system of the present application can recognize the gradual risk that cannot be found by the conventional threshold mode by constructing an individual activity context-adaptive evaluation reference and continuously quantizing the coordinated dynamic deviation of multi-dimensional physiological data, and can maintain the accuracy of the evaluation in different activity states.

[0041] Example 3: This example combines Figs. 1 to 3 to illustrate a remote central monitoring system for community medical treatment, such as Fig. 1As shown, the process starts from the user sensor as an external entity, which generates raw multidimensional physiological data and activity information, which are input to the data acquisition and preprocessing stage, in which the system filters the data based on the signal quality index SQI, discards low-quality data points, and then transmits the filtered and integrated data to the real-time statistical feature calculation stage, which simultaneously uses the K-means clustering algorithm to identify the current activity state to generate a real-time statistical feature set, which is then used to calculate the physiological stability index, the calculation benchmark of which is a certain individualized physiological baseline sub-model in a state-dependent baseline model cluster. The output of the calculation process is the physiological stability index, which is used to dynamically sort and present information to the medical staff as an external entity, and is also fed back to adaptively update the baseline model to generate an updated baseline model to maintain the baseline model cluster. In the process of calculating the physiological stability index, the system also generates a deviation vector to represent the deviation pattern by comparing it with the prototype vector in the prototype library, and adds a pattern label to the final output of the physiological stability index. At the same time, the specific method of calculating the deviation degree is to calculate the deviation degree by the Mahalanobis distance formula.

[0042] As shown in Fig. 2 The horizontal axis of the graph is time, from T=0h to T=72h, and the vertical axis corresponds to heart rate and blood oxygen saturation, respectively. According to the legend, the solid dot curve represents the mean heart rate, which increases steadily from 75 beats per minute at T=0h to nearly 90 beats per minute at T=72h. The dashed triangle curve represents the blood oxygen saturation, which decreases from 98.0% at T=0h to below 96.0% at T=72h. The dotted square curve represents the heart rate standard deviation, which decreases gradually over time, similar to the blood oxygen saturation. This series of coordinated data trends intuitively presents a gradual decline in the stability of the physiological system over time.

[0043] As shown in Fig. 3As shown, the architecture is divided into two parts, user-side edge computing device and cloud server side. On the user side, it contains an active state sensor three-axis acceleration sampling at a frequency of 50 Hz and a physiological sensor collecting heart rate, blood oxygen, signal quality index and other data. The data of the two are sent to the data acquisition and integration module for reception and screening, and the screening standard is SQI>0.5. The processed data enters the core physiological state analysis engine. The engine performs four functions: state-dependent baseline model cluster construction, identification of rest, light activity and moderate activity through K-means clustering, real-time state assessment using a 5-minute sliding window and Mahalanobis distance calculation to generate a physiological stability index between 0 and 1, deviation mode representation, and baseline model adaptive evolution when the stability threshold of 0.9 is triggered, with a weighted update of 95:5. The output of the engine, the physiological stability index stream + mode label, is sent to the cloud server side through encrypted transmission. The monitoring information presentation module pseudonymizes the received multi-user data and performs dynamic risk sorting, giving priority to those with lower physiological stability index. The monitoring interface displays real-time risk sorting list, mode label, historical change curve and original physiological data, and finally provides risk assessment and focused clinical intervention support for the medical staff terminal.

[0044] Example 4: This example describes an engineering procedure for individual calibration and key parameter determination of a remote central monitoring system to ensure that the internal model and decision threshold of the system are standardized before it is put into long-term monitoring. When a new user first uses the system, the system enters a 72-hour initial data acquisition and model calibration period. During this period, the data acquisition and integration module continuously records three-axis acceleration data at a sampling rate of 50 Hz, and synchronously acquires multi-dimensional physiological data of heart rate and oxygen saturation. The calibration procedure first determines the sliding time window length for real-time analysis. This parameter is set to balance the stability of data statistics and the immediacy of system response. Through analysis of the initial data set, the statistical characteristics calculated in a window shorter than 3 minutes show large fluctuations, while in a window longer than 10 minutes, the effective capture of state changes may be delayed. Therefore, this example determines 5 minutes as the working period of the sliding time window. The next step of the procedure is to construct the state-dependent baseline model cluster. The system applies the K-means clustering algorithm to process the 72-hour acceleration data. The feature vector used for clustering is composed of the mean and variance of the three-axis acceleration synthesis value in each 5-minute sliding time window. According to the silhouette coefficient analysis, when the number of clusters is greater than 10, the silhouette coefficient of each cluster is less than 0.5, indicating that the clustering result is not satisfactory. Therefore, this example determines 10 as the number of clusters in the baseline model cluster. When the number of data points is equal to 3, the data points form three clusters with clear boundaries and compact interiors, thereby objectively dividing the user's activity state into rest, light activity and moderate activity; then, the system filters out the corresponding multi-dimensional physiological data segments for the three identified activity states, and calculates an independent individualized physiological baseline sub-model for each activity state, the construction of the sub-model not only includes calculating the mean of each physiological parameter to form a baseline vector, but also includes calculating a covariance matrix representing the fluctuation and coupling relationship between the physiological parameters in this activity state, which is fixed in the model together with the baseline vector as the basis for calculating the deviation degree using the Mahalanobis distance formula.

[0045] Then, the procedure defines the prototype library required to characterize the deviation mode, which contains a set of standardized prototype vectors, each of which is consistent with the feature vector of the individualized physiological baseline sub-model in dimension; taking a feature space containing heart rate mean, blood oxygen saturation mean and heart rate standard deviation as an example, a prototype vector for characterizing psychogenic stress mode is set as a unit vector , the positive component 0.8 in the vector corresponds to the increase of heart rate mean, the negative component -0.2 corresponds to the slight decrease of blood oxygen saturation, and the negative component -0.57 corresponds to the decrease of heart rate variability, and the direction of the vector points to a specific physiological deviation mode in the multi-dimensional space; finally, the procedure calibrates the key decision threshold in the system operation; for the quality threshold of the signal quality index, by analyzing the signal quality index (SQI) distribution corresponding to the user's clear rest state and significant limb activity in the calibration period, the value 0.5 that can effectively distinguish the two distributions is set as the quality threshold; for the stability threshold used to trigger the adaptive evolution of the baseline model, the system uses the baseline model cluster completed by the initial calibration to backtest its own data in the calibration period, generates a distribution histogram of the initial physiological stability index, and sets the 90th percentile value of the distribution, such as 0.9, as the stability threshold, which can ensure that only when the user's physiological state is in its own high stability interval, the system will collect data for model updating; by executing the above procedure, all core models, algorithm parameters and decision thresholds in the system are determined based on the user's own initial data, thereby completing the individualized configuration.

[0046] Embodiment 5: This embodiment illustrates the fault-tolerant processing procedure of the remote central monitoring system in the face of a typical working condition of sensor data stream interruption, as well as the information processing method adopted to protect user data security; during the system monitoring period, if a user temporarily removes the physiological sensor due to a specific activity, resulting in the data acquisition and integration module failing to receive valid data points from the sensor within a preset time period of more than one sliding time window (5 minutes), the physiological state analysis engine deployed in the user-side edge computing device will automatically enter the data interruption state; in this state, the engine suspends sending the numerical physiological stability index to the cloud, and instead sends a pre-defined state code representing sensor signal loss; after receiving the state code, the monitoring information presentation module deployed in the cloud server executes the continuous automatic dynamic reordering logic, which will keep the user's last position before signal loss, while updating the user's display state to a specific visual identifier on the monitoring interface and adding a data interruption mode label next to their name.

[0047] To ensure data privacy in the process of centralized monitoring of multiple users, the system performs data isolation and pseudonymization processing in the data transmission and processing link from the user-side edge computing device to the cloud monitoring information presentation module; on the user side, after calculating the credibility-weighted physiological stability index, the physiological state analysis engine does not directly associate it with the user's real identity information for uploading, but first associates the index with a pseudonymized identifier generated by hashing the device unique identifier and the current timestamp, which does not contain any personally identifiable information; then, this data packet containing only the pseudonymized identifier and the physiological stability index value is sent to the cloud server via an encrypted channel; on the cloud side, the system maintains two isolated databases, one database stores the correspondence between the pseudonymized identifier and the user's real identity information, and is set with strict access permissions, and the other database only stores the pseudonymized identifier and its corresponding physiological stability index stream without real identity information; when performing dynamic risk sorting, the monitoring information presentation module only operates on the second database, and only when medical personnel actively request to view the detailed information of a specific user according to their permissions, the system will query the first database through the pseudonymized identifier to obtain the corresponding user identity information; this processing method makes the core data processed on the cloud de-identified information under normal operating conditions.

[0048] Example 6: This example illustrates a standardized engineering procedure for verifying the quality of the generated individualized physiological baseline model and finalizing the key algorithm parameters after system deployment and before entering the formal monitoring period; in the final verification phase of a system deployment, after the 72-hour initial data collection and model calibration period, the physiological state analysis engine will perform a baseline model quality self-check procedure before putting the newly created state-dependent baseline model cluster into use; the procedure aims to confirm that the data used for modeling comes from a stable physiological period, and the specific steps are to extract all multi-dimensional physiological data identified as resting state during the calibration period, and calculate the key parameters such as heart rate and blood oxygen saturation, and the coefficient of variation (CV) of all 5-minute sliding time window mean sequences, if the coefficient of variation of any parameter exceeds the pre-set quality control threshold of 15%, the system determines that the user's state fluctuates too much during the initial calibration period, and the generated baseline model is not reliable, at this time the system will automatically extend the calibration period until the data that meets the stability requirements is collected, this procedure is used to avoid the risk of baseline model misalignment caused by the user's unstable physical state in the initial deployment phase.

[0049] In this verification phase, the system also solidifies the calculation logic of the signal quality index (SQI), for signals derived from wrist photoplethysmography (PPG) sensors, the calculation of the signal quality index is determined as the following algorithmic steps: first, identify all PPG signal peaks in a 2-second sliding data window; second, calculate the standard deviation of the time interval (i.e. heart beat period) between all adjacent peaks in the window, and the standard deviation of all peak amplitudes; third, the value of the signal quality index is calculated as the inverse of the weighted sum of the two standard deviations, a low standard deviation value corresponds to a signal quality index close to 1; this specific algorithmic path converts the evaluation of signal quality from a qualitative description to an objectively reproducible calculation process; at the same time, for the baseline model adaptive evolution mechanism in the implementation, the update weight in the weighted average formula, i.e. the model evolution rate , is determined based on the technical trade-off between the model's tracking sensitivity to real baseline drift and its anti-interference ability to short-term physiological fluctuations; simulation tests show that a higher value of 0.1 will cause the model to overreact to short-term discomfort reactions lasting for several days, while a lower value of 0.01 may cause the model to fail to keep up with real physiological baseline changes occurring in the user over a period of weeks, this example finally adopts As the set value of this parameter, so that the model can effectively filter the physiological state fluctuations lasting less than a week, while being able to effectively track the long-term steady-state evolution in months; after completing all the above verification and confirmation steps, the remote central monitoring system of this user is confirmed to have completed deployment and calibration.

[0050] To further verify the necessity and superiority of the state-dependent baseline model cluster in the technical scheme of the present application, the following comparative examples are set to compare and illustrate.

[0051] Comparative Example 1: This comparative example aims to verify the key role of the state-dependent baseline model cluster in the present application in dealing with physiological parameter fluctuations caused by user daily activities. The same calculation simulation system and test data stream as in the second stage of Example 2 are used in this comparative example. The data stream simulates a healthy steady-state user whose activity state switches between rest and light activity. The only difference between the technical scheme of the monitoring system in this comparative example and the test group of Example 2 is that the monitoring system in this comparative example does not have the ability to construct a state-dependent baseline model cluster. Instead, it uses conventional technical ideas in the field to construct a single static individualized physiological baseline model based on multi-dimensional physiological data collected from the user in a resting state. During the monitoring period, regardless of the user's activity state, the system uses this single resting baseline model as the only judgment criterion to calculate the physiological stability index. In different activity states, the physiological stability index output results of the system in this comparative example are compared with those of the test group of Example 2 as shown in the following table.

[0052] Table 3: Comparison table of test data under activity state switching scenarios.

[0053]

[0054] As can be seen from the data in Table 3, when the simulated activity state is switched from rest to mild activity, the user's simulated heart rate physiologically rises from 75 beats per minute to 110 beats per minute. At this time, since the system of the comparative example only uses a single resting baseline model as the determination reference, the physiological heart rate increase caused by the activity is judged as a serious deviation from the resting steady state, and a low physiological stability index of 0.45 is output, resulting in a false risk prompt. In contrast, the test group system of Example 2 of the present application can automatically call a baseline sub-model corresponding to mild activity as the determination reference according to the activity state information, so the physiological stability index calculated by the system still maintains a high level of 0.92, accurately reflecting the user's physiological stability in this activity state. The results of this comparative test show that when the state-dependent baseline model cluster and the corresponding dynamic calling mechanism of the present application are absent, the conventional technical solution of using only a single static baseline model cannot effectively distinguish between physiological activity state changes and pathological state deviations. When the user is performing daily activities, the normal increase in parameters such as heart rate will result in an incorrect low stability index output, leading to a large number of false risk prompts, making it not practical in uncontrolled community monitoring scenarios.

[0055] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application.

[0056] Finally, it should be noted that the above examples are only used to illustrate the technical solutions of the present application and are not limiting. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application.

Claims

1. A community-oriented medical remote central monitoring system, characterized by, The method comprises: receiving activity state information and multi-dimensional physiological data from a user via an activity state sensor and at least one physiological sensor, respectively; identifying a plurality of typical activity states of the user based on the activity state information; calculating a set of statistical features for each of the typical activity states based on the multi-dimensional physiological data collected under the corresponding typical activity state, to construct a set of individualized physiological baseline sub-models corresponding to the plurality of typical activity states, respectively, the set of statistical features comprising the mean value of each dimension of the multi-dimensional physiological data, the standard deviation of each dimension of the multi-dimensional physiological data, and the correlation coefficient between any two dimensions of the multi-dimensional physiological data, the set of individualized physiological baseline sub-models collectively forming a state-dependent baseline model cluster; and selecting, from the state-dependent baseline model cluster, an individualized physiological baseline sub-model corresponding to the current activity state of the user as a decision reference, according to the current activity state of the user monitored in real time during a monitoring period; continuously collecting real-time multi-dimensional physiological data in a sliding time window, and calculating a set of real-time statistical features identical in dimension and type to the set of statistical features of the individualized physiological baseline sub-model; calculating a deviation degree between a real-time state vector defined by the set of real-time statistical features and a baseline vector defined by the set of statistical features of the decision reference by a preset distance formula, and converting the deviation degree into a continuously changing physiological stability index; receiving physiological stability indices from a plurality of users, and continuously and automatically dynamically reordering the positions of the plurality of users on a monitoring interface according to the numerical values of the physiological stability indices; and continuously monitoring the physiological stability index generated by the physiological state analysis engine, and automatically triggering a baseline model updating process when the physiological stability index is continuously greater than a preset stability threshold for a preset period of time, and calculating a candidate baseline model using the multi-dimensional physiological data collected under the corresponding activity state during the period of time when the physiological stability index is continuously greater than the stability threshold. fusing the candidate baseline model with the corresponding individualized physiological baseline sub-model in the state-dependent baseline model cluster by a weighted average formula; and The physiological state analysis engine is further configured to perform the following rules: after generating the physiological stability index, determine a deviation vector pointing from the baseline vector to the real-time state vector; perform a directional comparison of the deviation vector against a prototype library containing a plurality of prototype vectors representing directions of underlying physiological disturbances to identify a prototype vector having a highest directional similarity with the deviation vector; wherein the directional similarity The directional similarity is calculated by the following equation: wherein, is the deviation vector, is any prototype vector in the prototype library, • is the vector dot product operation, is the Euclidean norm of a vector; and based on the identified prototype vector, attach a pattern label representing the deviation pattern to the physiological stability index.

2. The remote central monitoring system for community medical treatment according to claim 1, wherein, acquiring a signal quality index representing the quality of the output signal of the at least one physiological sensor by the data acquisition and integration module, and calculating the average value of the signal quality indices of all non-discarded data points in the sliding time window as a credibility factor when generating the physiological stability index, wherein the multi-dimensional physiological data points with a signal quality index lower than a preset quality threshold are discarded and not used to calculate the set of real-time statistical features in the sliding time window; and multiplying the physiological stability index by the credibility factor to generate a credibility-weighted physiological stability index.

3. The remote central monitoring system for community healthcare according to claim 1, wherein, The stability threshold is set to 0.9; the weighted average formula is: updated individualized physiological baseline sub-model = 95% * old individualized physiological baseline sub-model + 5% * candidate baseline model.

4. The remote central monitoring system for community healthcare of claim 1, wherein, The distance formula used by the physiological state analysis engine when calculating the deviation degree is the Mahalanobis distance formula.

5. The remote central monitoring system for community healthcare of claim 1, wherein, The monitoring information presentation module is configured to perform continuous automatic dynamic reordering, specifically, users with lower physiological stability index values are displayed in a higher priority position on the monitoring interface.

6. The remote central monitoring system for community healthcare of claim 1, wherein, The physiological state analysis engine is deployed in a user-side edge computing device; the edge computing device is configured to send a data stream mainly carrying the physiological stability index to the monitoring information presentation module deployed in the cloud server.

7. The remote central monitoring system for community healthcare of claim 1, wherein, The physiological state analysis engine is configured to identify multiple typical activity states of the user based on the activity state information, specifically: using an unsupervised K-means clustering algorithm to process the three-axis acceleration data stream output by the activity state sensor to automatically cluster the activity data of the user into three typical activity states of rest, light activity and moderate activity.

8. The remote central monitoring system for community healthcare of claim 1, wherein, The monitoring information presentation module is also configured to display the mode label corresponding to each user's physiological stability index value on the monitoring interface, and when a specific user is selected, the physiological stability index historical change curve and the original multi-dimensional physiological data curve of the user are retrieved and presented.

Images