Health monitoring management system based on multi-modal indexes

Abstract

The invention discloses a health monitoring management system based on a multi-modal index, and belongs to the technical field of health monitoring, and the system specifically comprises a data collection unit which synchronously collects physiological signals at the same anatomical position of a user through a professional medical device and a wearable device; the calibration model construction unit is used for extracting reference waveform characteristics of professional medical equipment, analyzing a signal offset mode of the wearable equipment, fitting a dynamic correction curve and generating a calibration comparison table; the dynamic correction unit is used for inputting the real-time data of the wearable equipment into a calibration comparison table for correction, generating a reconstructed data stream aligned with the time domain of the reference signal, and ensuring the synchronization of waveform key points; the calibration verification unit is used for calling high-precision data to verify the calibration result in a preset period, and updating the calibration rule base when the matching error of a plurality of continuous calibration windows is lower than a threshold value; according to the method, the wearable device is periodically calibrated through the high-precision professional medical device, and the daily health monitoring precision of the user is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of health monitoring, in particular to a health monitoring management system based on multi-modal indexes. BACKGROUND

[0002] With the explosive development of wearable devices and Internet of Things technology, remote health monitoring systems based on multi-modal physiological signal fusion are accelerating the penetration into chronic disease management, home intelligent nursing and sports health management scenarios. Such systems synchronously collect heart rate variability, 12-lead ECG waveform, blood oxygen saturation, skin conductance and other multi-dimensional physiological indicators, building a dynamic monitoring network covering users' basic vital signs, cardiovascular function and stress state. They not only can realize real-time early warning of abnormal heart rate, but also can analyze the progress trend of chronic diseases through long-term data accumulation.

[0003] However, consumer-grade wearable devices are generally limited by terminal size, manufacturing cost and battery life, and commonly use low-cost sensors, simplified signal conditioning circuits and low-power sampling chips. Compared with professional medical-grade devices, consumer-grade devices have significant gaps in sampling accuracy and signal processing capability, resulting in a large amount of noise mixed in long-term monitoring data, which may mask the subtle changes of early pathological signals, making it difficult for health trend analysis based on wearable devices to meet clinical application standards.

[0004] Therefore, how to improve the monitoring accuracy of consumer-grade devices through cross-device data calibration mechanism without significantly increasing the cost and complexity of the terminal. SUMMARY

[0005] The present application aims to provide a health monitoring management system based on multi-modal indexes, which solves the following technical problems:

[0006] How to improve the monitoring accuracy of consumer-grade devices through cross-device data calibration mechanism without significantly increasing the cost and complexity of the terminal.

[0007] The purpose of the present application can be achieved by the following technical solutions:

[0008] A health monitoring management system based on multi-modal indexes, comprising:

[0009] A data acquisition unit for synchronously acquiring physiological signals at the same anatomical position of the user by professional medical devices and wearable devices. The hardware clock of the two types of devices is synchronized at the start of each calibration period, and the raw data collected is marked to the millisecond of the acquisition time and the device accuracy level.

[0010] The calibration model construction unit is configured to extract reference waveform features of the professional medical equipment, analyze a signal offset mode of the wearable device, fit a dynamic correction curve according to a variation trend of an amplitude deviation amount, and generate a calibration table containing time-environment-equipment three-dimensional parameters;

[0011] The dynamic correction unit is configured to input real-time data of the wearable device into the calibration table to perform signal correction, and generate reconstructed data stream aligned with a time domain of the reference signal, wherein waveform key points of the reconstructed data stream are time-synchronized with R-wave peaks and T-wave end points of the reference signal.

[0012] The calibration verification unit is configured to call professional medical equipment data to verify calibration results of the wearable device at a preset period, update a calibration rule library when matching errors of a continuous set number of calibration windows are lower than a threshold value, and record an effective time period and retain historical version data in the calibration rule library.

[0013] As a further scheme of the present application, the data acquisition unit specifically includes:

[0014] The professional medical equipment and the wearable device set signal acquisition points at the same anatomical position of a user, and contact pressures of sensor surfaces of the two types of devices with the skin are maintained consistent through a micro pressure feedback device.

[0015] The two types of devices perform hardware clock synchronization when starting each calibration period, and align internal timers through physical connection or a near field communication protocol; the synchronized devices continuously acquire the same type of signals within a complete physiological period, and the complete physiological period contains at least one complete heartbeat activity and a breathing cycle.

[0016] The acquired original data is marked to an acquisition time point accurate to milliseconds, and is labeled with a device precision level identifier; the reference signal acquired by the professional medical equipment is stored in an independent storage area, and the independent storage area is physically isolated from a to-be-calibrated signal storage area; the to-be-calibrated signal of the wearable device is transmitted to the dynamic correction unit in real time, and the transmission process retains quantization precision and a sampling interval of the original signal.

[0017] As a further scheme of the present application, in the calibration model construction unit, the extraction process of the reference waveform features is as follows:

[0018] The continuous heartbeat signal of the professional medical equipment within a complete calibration period is intercepted, and the R-wave peak amplitude and the T-wave end point slope of each heartbeat period are identified; the original signal of the wearable device within the same period is cut according to the heartbeat period, and the time deviation amount of the R-wave detection point is used as an offset reference during the cutting.

[0019] The analysis of the signal offset mode includes construction of a statistical histogram of the amplitude deviation amount and calculation of a sliding window mean value of the time delay amount; the fitting of the dynamic correction curve is based on time series data of the amplitude deviation amount, and the time series data is divided into intervals according to each hour as a unit.

[0020] The generation of the calibration table contains an interpolation compensation term of the environmental temperature parameter, and the interpolation compensation term is obtained according to temperature sensor historical data regression analysis; the rolling update of the calibration table is realized by replacing the earliest historical version, and the historical version data is stored with a snapshot of the environmental parameter at the time of version generation.

[0021] As a further scheme of the present application, the analysis process of the signal offset mode is:

[0022] The original signal of the wearable device is divided into independent segments according to the heartbeat period, and the R-wave peak point is taken as the cycle starting point during the division; each independent segment is superimposed with the corresponding period reference signal of the professional medical device, and the absolute time stamp of the R-wave peak point is aligned during the superposition; the amplitude difference area of the superimposed waveform is calculated, and the amplitude difference area is the integral difference value of the two waveforms in the vertical direction;

[0023] The time delay cumulative amount is calculated, and the time delay cumulative amount is the sum of the time offsets of the two waveforms at the same feature point; the fixed deviation component caused by the inherent characteristics of the device in the amplitude difference area is identified, and the fixed deviation component is obtained through historical data statistical analysis; the random fluctuation component caused by the environmental temperature fluctuation is separated, and the random fluctuation component is extracted through a sliding window filtering algorithm; the fixed deviation component is written into the calibration rule as a basic correction amount, and the random fluctuation component generates a dynamic compensation term associated with the environmental temperature.

[0024] As a further scheme of the present application, the signal correction process of the dynamic correction unit includes:

[0025] A combination of correction parameters matching the current time stamp, environmental temperature and device wearing state is retrieved from the calibration table; a baseline reset operation is performed on the original signal of the wearable device, and the baseline reset refers to the average amplitude of the reference signal in the same time period; an amplitude correction coefficient and a time delay compensation amount are applied, and the correction coefficient is weighted according to the frequency component of the signal; the waveform splicing technology is used to generate the reconstructed data stream, and the R-wave vertex of the reference signal is taken as the time alignment anchor point during splicing;

[0026] During the time domain alignment process, the sampling points of the wearable device signal are interpolated and resampled to make the time resolution consistent with that of the reference signal; the QRS complex width of each heartbeat period in the reconstructed data stream is controlled to be within twice the sampling interval difference from the corresponding value of the reference signal; the corrected signal stream is transmitted to the calibration verification unit in real time, and the calibration table version identification currently used is attached during the transmission.

[0027] As a further scheme of the present application, in the calibration verification unit, the matching process of the calibration window is:

[0028] The calibration window is automatically triggered to open when the user is in a resting state and the ambient temperature fluctuation is less than a set range; the duration of the calibration window covers at least twenty consecutive heartbeat periods, ensuring that the signal characteristics in different heart rate states are included; the calculation of the matching error uses a dynamic time warping algorithm, and local path constraints are set in the algorithm to exclude accidental deviations; the error threshold is dynamically adjusted according to the measurement accuracy of professional medical equipment, and the adjustment range is related to the technical specifications of the equipment;

[0029] The update of the calibration rule library includes new entries and replacement of historical entries, and the generalization ability of new entries in historical data sets is verified during replacement; the trend analysis of historical version data calls the moving average algorithm to identify the drift law of the calibration parameters with the length of device use; the trend analysis result is used to generate optimization suggestions for the calibration control table, and the optimization suggestions include pre-adjustment of compensation coefficients.

[0030] As a further scheme of the present application, the dynamic adjustment process of the error threshold is:

[0031] The signal stability indicators of the professional medical equipment within the last twenty-four hours are obtained, including amplitude standard deviation and time delay jitter value; the basic error tolerance is calculated according to the stability indicators, and the basic error tolerance is proportional to the standard deviation; a device usage time decay factor is superimposed, and the decay factor increases linearly with the increase of the cumulative working time of the device;

[0032] The final error threshold is set as the product of the basic error tolerance and the decay factor, and the product result is limited within the maximum allowed error range of the device; when the matching error exceeds the final threshold, the version rollback mechanism of the calibration rule library is triggered; the version rollback mechanism retrieves the last three valid versions from the archive storage area, and selects the historical version closest to the current environmental parameters; after the rollback operation is completed, the calibration verification process is restarted and the current invalid rule entry is disabled.

[0033] As a further scheme of the present application, the historical version data specifically includes:

[0034] During the daily system maintenance window, data cleaning is performed on the historical versions, and invalid parameter combination entries are removed during cleaning; valid entries are stored in classified storage according to environmental temperature gradient, and the classification interval is a set temperature value, and there is an overlapping buffer zone between adjacent classification intervals;

[0035] When the trend analysis calls the historical version, the data in the same temperature classification and continuous time period is preferentially used; the drift law of the calibration parameters is fitted by the least square method to obtain the slope of the straight line, and the slope value is used to predict the correction increment of the next period; the prediction result is converted into a pre-adjustment instruction of the calibration control table, and the pre-adjustment instruction is automatically loaded at the beginning of the next calibration period;

[0036] The application of the pre-adjustment instruction needs to pass through a temporary verification process, the temporary verification adopts a shortened calibration window to quickly check the prediction rationality, the pre-adjustment instruction that passes the check is merged into the formal calibration rule, and a prediction optimization mark is marked in the version identification.

[0037] As a further scheme of the application: when it is detected that the signal amplitude of the wearable device continuously exceeds the device range, an emergency calibration mode is enabled; the emergency calibration mode forcibly calls the latest reference signal of the professional medical device, and skips the regular calibration verification process; the interception period of the reference signal is extended forward to before the start point of the abnormal data, ensuring that the normal signal characteristics are covered;

[0038] A bidirectional interpolation method is used when reconstructing the data, and the missing waveform is speculated based on the normal data before and after the abnormal period; the data segment after interpolation reconstruction is marked as temporary calibration data, and the temporary data is converted into formal data after passing the supplementary calibration verification;

[0039] The supplementary calibration verification adopts a high-density sampling mode, and the reference data quantity of the high-density sampling is increased to three times of that of the regular mode; the temporary data that passes the verification triggers the priority update of the calibration rule library, and the effective period of the new rule item is shortened to a set proportion of the original value.

[0040] The beneficial effects of the application are:

[0041] The application effectively solves the data deviation problem between the wearable device and the professional medical device by constructing a closed-loop system of multi-modal data synchronous acquisition, dynamic calibration and continuous verification, and significantly improves the accuracy and reliability of daily health monitoring. Through hardware clock synchronization and pressure feedback control, multi-device signal synchronous acquisition at the same anatomical position is realized, and the clock deviation and time domain misplacement problems are eliminated; based on the three-dimensional parameter calibration reference table and the dynamic correction curve, the signal drift caused by device aging, temperature fluctuation and wearing state change is compensated in real time, and the adaptability of the calibration model is improved; the R wave peak alignment and time domain interpolation technology are used to ensure that the reconstructed data stream and the reference signal are strictly synchronized in terms of waveform key points, and the QRS complex width difference is controlled within a very small range; a periodic calibration verification mechanism is established, the matching error is calculated through the dynamic time warping algorithm, and the threshold is dynamically adjusted according to the device stability index to realize adaptive update of the calibration rule; an emergency calibration mode is designed for abnormal signals, combined with bidirectional interpolation and high-density sampling verification, to quickly restore data reliability and avoid health warning delay caused by signal fluctuation. Through the above technical means, the monitoring error of the wearable device is significantly reduced, close to the level of professional medical devices, and more reliable data support is provided for remote health management. BRIEF DESCRIPTION OF DRAWINGS

[0042] The application will be further described below with reference to the drawings.

[0043] Figure 1It is a system structure schematic diagram of the present application. DETAILED DESCRIPTION

[0044] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0045] Please refer to Figure 1 The present application is a health monitoring management system based on multi-modal indicators, which comprises:

[0046] The system first realizes the synchronous acquisition of multi-source data through the data acquisition unit. The professional medical equipment and the wearable device arrange signal acquisition points at the same anatomical position of the user, and maintain the consistent contact pressure of the sensor and the skin through the micro pressure feedback device, thereby reducing the influence of physical differences on the signal. At the start of each calibration period, the two types of devices perform hardware clock synchronization through physical connection or near field communication protocol, to ensure the accurate alignment of the internal timers. After synchronization, the devices continuously acquire the same type of signal within a complete physiological cycle, and the acquired raw data is marked to the millisecond acquisition time and labeled with the device accuracy level. The reference signal acquired by the professional medical equipment is stored in an independent storage area, which is physically isolated from the calibration signal of the wearable device, to ensure data integrity.

[0047] The calibration model construction unit is responsible for establishing accurate calibration rules. This unit extracts the continuous heartbeat signals of the professional medical equipment within a complete calibration period, identifies the reference waveform features such as R-wave peak amplitude and T-wave end slope, and divides the wearable device raw signal according to the heartbeat period. By constructing the amplitude deviation amount statistical histogram, calculating the time delay amount sliding window mean value, analyzing the signal offset mode, and separating the device inherent deviation and the random fluctuations caused by environmental factors, a calibration control table containing time-environment-device three-dimensional parameters is generated based on the amplitude deviation amount time series data, and the dynamic adjustment of the calibration parameters is realized.

[0048] The dynamic correction unit inputs the real-time data of the wearable device into the calibration control table, performs baseline reset operation first, and then applies the amplitude correction coefficient and the time delay compensation amount for signal correction. Through waveform splicing technology, the R-wave vertex of the reference signal is used as the time alignment anchor point, the wearable device signal is interpolated and resampled, the reconstructed data stream is generated which is time-domain aligned with the reference signal, and the time synchronization of the waveform key points is ensured.

[0049] The calibration verification unit calls the professional medical equipment data verification calibration result at a preset period. The calibration window is opened when the user is at rest and the environment is stable, the matching error is calculated by using a dynamic time warping algorithm, and the error threshold is dynamically adjusted according to the device stability index. When the matching errors of a plurality of continuous calibration windows are lower than the threshold, the calibration rule library is updated, the effective time period is recorded and the historical version data is retained, and the continuous optimization of the calibration rule is realized.

[0050] In a preferred embodiment of the present application, the data acquisition unit specifically comprises:

[0051] The system accurately deploys the signal acquisition points of the professional medical equipment and the wearable equipment at the same anatomical position of the user's body, and constructs a closed-loop pressure adjustment system by using a miniature pressure feedback device. The device is built-in with a high-precision pressure sensor, which monitors the contact pressure between the sensor and the skin in real time at a frequency of 20 times per second, and drives a miniature servo mechanism through an adaptive adjustment algorithm to dynamically adjust the tightness of the equipment wearing, so as to ensure that the pressure during the acquisition of the two types of equipment always maintains within an error range of ±0.5N, effectively avoiding signal distortion and measurement deviation caused by pressure difference.

[0052] In terms of time synchronization, when each calibration period is started, the hardware clock synchronization mechanism is triggered between the devices through physical connection (such as USB interface, Type-C interface) or near field communication protocol (such as Bluetooth 5.0, NFC). The system first calibrates the internal timer of the device through a high-precision clock chip, and then compensates the error by using a bidirectional timestamp exchange algorithm. The algorithm calculates the clock offset and clock drift rate through multiple round-trip time measurements, achieving nanosecond-level clock synchronization accuracy. After synchronization, the devices take the synchronized clock as the reference, and synchronously collect the same type of physiological signals such as electrocardiogram and pulse within 3 complete physiological cycles at a sampling frequency of 200Hz, ensuring that the collected data covers at least one complete heartbeat cycle and breathing cycle, providing sufficient data samples for subsequent analysis.

[0053] The collected raw data are all marked with time stamps accurate to milliseconds and attached with device precision level identifiers. The reference signals collected by the professional medical equipment are stored in an independent physically isolated storage area through a special data channel. The storage area is designed with double backup redundancy and is equipped with a data encryption module to prevent data tampering or leakage; while the signals to be calibrated of the wearable equipment are directly transmitted to the dynamic correction unit at the original quantization precision and sampling interval through the real-time transmission protocol, and error correction coding technology is used in the transmission process to ensure zero loss and zero distortion of data.

[0054] In another preferred embodiment of the present application, in the calibration model construction unit, the extraction process of the reference waveform features is:

[0055] Baseline waveform feature extraction stage: The system first intercepts the continuous heartbeat signal of the professional medical equipment in the complete calibration cycle, uses the adaptive threshold algorithm combined with the second derivative zero-crossing detection technology to accurately identify the R-wave peak position of each heartbeat cycle, and calculates its amplitude value. At the same time, through the T wave form analysis algorithm, the T wave end position is determined based on the waveform curvature rate, and then the slope of the point is calculated. For the original signal of the wearable device in the same cycle, the system takes the R-wave time point detected by the professional medical equipment as the reference, and aligns the time domain by calculating the time deviation. In specific implementation, the dynamic time warping (DTW) algorithm is used to calculate the optimal mapping path between the two signals to obtain the time offset of the R-wave detection point, and the offset is used as the offset reference for cutting the low-precision signal to ensure that the signals of the two types of devices are strictly aligned in time domain.

[0056] Signal offset mode analysis stage: The system constructs a three-dimensional amplitude deviation statistical histogram, where the X-axis represents the time sequence, the Y-axis represents the amplitude deviation value, and the Z-axis represents the deviation frequency. Through the histogram, the deviation distribution characteristics in different time periods and different amplitude intervals can be intuitively presented. At the same time, a sliding window with a length of 20 heartbeat cycles is used to track and calculate the time delay in real time. The mean value is updated once every cycle, effectively capturing the dynamic changes of signal transmission delay. To further explore the time distribution law of amplitude deviation, the system divides the time sequence data into intervals every hour, and performs spectral analysis on the deviation data in each interval to identify high-frequency fluctuation components and low-frequency trend components, providing a basis for the fitting of the subsequent dynamic correction curve.

[0057] Calibration control table generation and update stage: When generating the calibration control table, the system introduces an interpolation compensation term for the environmental temperature parameter. Specifically, by collecting historical data of the temperature sensor, a temperature-deviation regression model is established. The model uses a quadratic polynomial fitting method combined with least squares method to estimate the model parameters, realizing accurate quantification of the temperature influence. In actual application, according to the current temperature value, the compensation coefficient corresponding to the temperature value is calculated through the interpolation algorithm, and the compensation coefficient is integrated into the calibration parameter. The calibration control table adopts a rolling update mechanism, and each update retains the last 30 historical versions, and each version is attached with a snapshot of the environmental parameters at the time of generation, including temperature, humidity, air pressure and other multi-dimensional environmental data. When backtracking analysis is needed, the historical calibration version closest to the current environmental conditions can be quickly located according to the similarity of environmental parameters.

[0058] In a preferred case of the embodiment, the analysis process of the signal offset mode is as follows:

[0059] The system segments the raw signal of the wearable device by heart beat cycle, and each cycle starts with the R-wave peak point and intercepts a 300ms signal segment. The waveform of each segment is superimposed on the reference signal of the corresponding cycle of the professional medical device, and the absolute timestamp of the R-wave peak point is strictly aligned during superimposition. The integral difference of the two waveforms in the vertical direction is calculated by the numerical integration method to obtain the amplitude difference area. The area value directly reflects the overall deviation degree of the two waveforms in the entire heart cycle.

[0060] The system further calculates the time delay accumulation, that is, the sum of the time offset of the two waveforms at the same feature points (such as R-wave, T-wave, and P-wave). Through statistical analysis of a large amount of historical data, a device inherent characteristic model is established, and a fixed deviation component caused by the hardware characteristics of the device in the amplitude difference area is identified. The component remains relatively stable under different environmental conditions, and is therefore written into the calibration rule as a basic correction amount. For the random fluctuation component caused by the fluctuation of environmental temperature, the system uses an adaptive sliding window filtering algorithm to extract it. The algorithm dynamically adjusts the window size according to the temperature change rate, reduces the window size when the temperature changes sharply, and improves the response capability to rapid changes; when the temperature is relatively stable, the window is increased to enhance the filtering effect. The extracted random fluctuation component and the temperature data establish a mapping relationship to generate a dynamic compensation term, realizing real-time correction of the influence of environmental temperature.

[0061] In another preferred embodiment of the application, the signal correction process of the dynamic correction unit comprises:

[0062] The system first retrieves the matched correction parameter combination from the three-dimensional parameter index structure of the calibration table based on the current timestamp, real-time data of the environmental temperature sensor, and the output of the wearing state sensor (such as a three-axis accelerometer). The calibration table uses the hash index technology, takes the time (accurate to the hour), the temperature (±0.5℃ interval), the device number, and the wearing posture (rest / movement) as the key value, and ensures that the parameter retrieval time is less than 10ms. The baseline reset operation takes the average amplitude of the reference signal in the corresponding time period as the reference, eliminates the baseline drift of the low-precision signal through the high-pass filtering algorithm (cutoff frequency 0.5Hz), and aligns the direct current component of the corrected signal with the reference signal.

[0063] The amplitude correction and time delay compensation link adopts a frequency band processing strategy: the signal spectrum is divided into low frequency (<0.5Hz), medium frequency (0.5-30Hz) and high frequency (>30Hz) three frequency bands, and different correction coefficients are applied respectively. The low frequency band (mainly including slow varying signals such as respiratory wave) adopts linear gain compensation, the medium frequency band (including characteristic signals such as electrocardiogram QRS complex) is nonlinearly corrected by table lookup method, and the high frequency band (mainly for noise) is combined with band pass filtering and time delay correction. The waveform splicing technology takes the R wave vertex of the reference signal as the time anchor point, uses cubic spline interpolation to resample the low-precision signal, so that the time resolution is improved from the original 10ms to 1ms consistent with the reference signal, and the dynamic time warping algorithm is used to ensure that the width of the reconstructed QRS complex is within 2ms (i.e. twice the sampling interval) of the reference signal. The corrected data stream is attached with calibration control table version identification, time stamp and device state information, and is pushed to the calibration verification unit in real time by an encrypted transmission protocol.

[0064] In another preferred embodiment of the present application, in the calibration verification unit, the matching process of the calibration window is:

[0065] When triggering the calibration window to open, the system monitors the multi-source data in real time to ensure the effectiveness of the verification environment: through the fusion data of the acceleration sensor and the gyroscope, the support vector machine algorithm is used to judge whether the user is in a resting state, when the root mean square value of three-axis acceleration is continuously less than 0.1g for 5 minutes and the attitude angle change rate is less than 5° / s, it is determined that the user is in a resting state; at the same time, combined with the environmental temperature sensor data, if the temperature fluctuation is not more than ±1℃ within 10 minutes, the calibration window is automatically activated. The window time is accurately set to cover 20-30 consecutive heartbeat periods (about 2-3 minutes), during which the signals of wearable devices and professional medical devices are synchronously collected, to ensure that the waveform characteristics under different heart rate states such as sinus rhythm and tachycardia are covered.

[0066] The matching error calculation adopts an optimized dynamic time warping (DTW) algorithm, in order to avoid accidental deviation from interfering with the verification result, the local path constraint condition is set in the algorithm: the slope change of the time series matching path is limited to not more than ±1 sampling interval, and the path backtracking step number is limited to not more than 3 steps, to ensure that the waveform alignment is more consistent with the true form of the physiological signal. The error threshold is deeply bound with the technical parameters of the professional medical device, the system collects the signal stability indicators of the device within 24 hours in real time, the amplitude standard deviation reflects the signal fluctuation degree, and the time delay jitter value reflects the clock synchronization accuracy. Based on the formula "basic error tolerance = 1.5×amplitude standard deviation + 2×time delay jitter value", the reference value is calculated, and a device usage time decay factor (initial value is 1.0, increases by 0.1 every 100 hours of cumulative work) is introduced, the product of the two is truncated by the maximum allowed error boundary of the device (such as the heart rate measurement error is not more than ±2bpm), to form a dynamically adjusted error judgment standard.

[0067] The update of the calibration rule library follows a strict double-checking mechanism: the addition of a calibration entry requires a five-fold cross-validation of historical data sets, and its generalization error is verified to be lower than 5% under different user groups and environmental conditions; when replacing historical entries, the system automatically compares the correction effects of new and old rules in low (< 60 bpm), medium (60-100 bpm), and high (> 100 bpm) heart rate intervals, and retains the parameter combination with the best overall performance. For historical version data, the system uses a moving average algorithm (window length set to 7 days) to fit the drift curve of the calibration parameters with the use time, combines the least square method to predict the compensation coefficient increment in the next 24 hours, generates pre-adjustment instructions for the calibration control table in advance, and realizes preventive error correction.

[0068] When the matching error of three consecutive calibration windows exceeds the final threshold value, the calibration verification unit immediately starts the version rollback mechanism: the last three effective calibration versions are called from the archive storage area, a multi-dimensional feature vector is constructed based on 12 environmental parameters such as temperature, humidity, air pressure, and altitude, and the cosine similarity algorithm is used to select the historical version that best matches the current environmental conditions. After rollback, the system automatically enables the rapid verification process: temporarily increases the sampling frequency to 500 Hz, verifies the QRS complex morphology, RR interval, and other key waveform features at high density, disables the current invalid rule entry after verification, and triggers a new round of calibration model iteration to ensure the long-term reliability and accuracy of health monitoring data.

[0069] In another preferred embodiment of the present application, the historical version data specifically includes:

[0070] The historical version data management system builds a fine data processing and trend analysis system. During the daily system maintenance period (set as 2:00-4:00 system low peak period), the system automatically executes the data cleaning process: first, the isolated forest algorithm is used to identify the abnormal values in the calibration parameters, and the invalid items whose matching error exceeds twice the maximum allowed error of the device for three consecutive calibration periods are removed; then the valid items are stored according to the environmental temperature gradient classification, and the classification interval is set to 2℃ (such as 20-22℃, 22-24℃), and a 0.5℃ overlapping buffer band is set between adjacent intervals to ensure the continuity of data retrieval when the temperature fluctuates. When the trend analysis module calls historical data, the calibration records of the same temperature classification for 7 consecutive days are extracted first to construct a time series data set. The least square method is used to fit the change curve of the calibration parameters with the device usage time, and the slope of the straight line is calculated as the drift rate index, which is combined with the cumulative working time of the device to predict the correction increment of the next period. The prediction result is converted into a pre-adjustment instruction, which is automatically loaded into the calibration control table 5 minutes before the start of each calibration period. Before loading, a temporary verification process is performed: a shortened calibration window covering 5 heartbeat periods is opened, and the waveform matching error before and after pre-adjustment is compared. If the error change rate is less than 10%, it is determined to pass, and the verified instruction is merged into the formal rule, and a "P" (Prediction) mark is added after the version number to distinguish it.

[0071] In another preferred embodiment of the application, when the signal amplitude of the wearable device exceeds the device range for 3 seconds (such as the photoplethysmogram signal exceeding 95% of the ADC full range), the system immediately enables the emergency calibration mode. In this mode, the system forces the professional medical device to call the reference signal from 30 seconds before the anomaly to the current time, accurately locates the starting point of the anomaly through a sliding window detection algorithm, and ensures that the intercepted reference signal contains complete normal waveform features. The reconstructed abnormal data uses a bidirectional interpolation method: taking the previous and next 5 normal heartbeat periods as anchor points, combining cubic spline interpolation and waveform template matching technology, and dynamically generating a substitute waveform that meets the physiological characteristics. The reconstructed data segment is marked as temporary calibration data (suffix "T" after the version number), and the supplementary calibration verification process is immediately started. This process increases the sampling frequency to 3 times that of the normal mode (such as from 100Hz to 300Hz), and extends the collection time to 60 heartbeat periods. The verification is performed by calculating the shape similarity (requirement ≥0.95) and feature point deviation (≤15ms) between the reconstructed data and the reference signal. The verified temporary data triggers the priority update of the calibration rule library, and the effective time of the newly generated rule entry is shortened to 50% of the normal rule (such as from 24 hours to 12 hours), to quickly respond to device sudden anomalies, while retaining a sufficient time window to observe the rule effect. The entire emergency calibration process is completed within 20 seconds, ensuring the continuity and reliability of health monitoring data.

[0072] The above has been described in detail one embodiment of the present application, but the content is only the preferred embodiment of the present application, cannot be considered for limiting the scope of the present application. Any equivalent changes and improvements made in the scope of the present application, should still belong to the scope of the present application.

Claims

1. A health monitoring and management system based on multimodal indicators, characterized in that: include: The data acquisition unit is used to synchronously collect physiological signals from the user's anatomical location using professional medical equipment and wearable devices. The two types of devices synchronize their hardware clocks at the start of each calibration cycle. The collected raw data is marked with the collection time and device accuracy level accurate to the millisecond; The calibration model construction unit is used to extract the baseline waveform characteristics of professional medical equipment, analyze the signal offset pattern of wearable devices, fit the dynamic correction curve according to the change trend of the amplitude deviation, and generate a calibration comparison table containing the three-dimensional parameters of time, environment, and equipment; A dynamic correction unit is used to input the real-time data of the wearable device into the calibration table for signal correction, generating a reconstructed data stream that is aligned with the reference signal in the time domain. The waveform key points of the reconstructed data stream are synchronized with the R wave peak and T wave end point of the reference signal. The calibration verification unit is used to call professional medical equipment data at a preset period to verify the calibration results of the wearable device. When the matching error of a set number of consecutive calibration windows is lower than the threshold, the calibration rule library is updated. The calibration rule library records the effective time period and retains historical version data.

2. A health monitoring and management system based on multimodal indicators according to claim 1, characterized in that: The data acquisition unit specifically includes: Professional medical devices and wearable devices set up signal collection points at the same anatomical position of the user, and the contact pressure between the sensor surface and the skin of the two types of devices is maintained consistent through a micro pressure feedback device; The two types of devices synchronize their hardware clocks at the start of each calibration cycle, aligning their internal timers via a physical connection or near-field communication protocol. The synchronized devices continuously collect the same type of signals over a complete physiological cycle, which includes at least one complete heartbeat and respiratory cycle. The collected raw data is marked with the collection time accurate to milliseconds and annotated with the equipment accuracy level identification; the reference signal collected by professional medical equipment is stored in an independent storage area, which is physically isolated from the storage area of ​​the signal to be calibrated; the signal to be calibrated of the wearable device is transmitted to the dynamic correction unit in real time, and the transmission process retains the quantization accuracy and sampling interval of the original signal.

3. The health monitoring and management system based on multimodal indicators according to claim 1, characterized in that: In the calibration model construction unit, the process of extracting the reference waveform features is as follows: Intercept the continuous heartbeat signal of professional medical equipment within a complete calibration cycle, identify the R wave peak amplitude and T wave endpoint slope of each heartbeat cycle; split the original signal of the wearable device within the same cycle according to the heartbeat cycle, and use the time deviation of the R wave detection point as the offset benchmark during segmentation; The analysis of signal deviation patterns involves constructing a statistical histogram of amplitude deviation and calculating a sliding window mean of delay. The dynamic correction curve is fitted based on the time series data of amplitude deviation, which is divided into hourly intervals. The generation of the calibration reference table includes interpolation compensation items for ambient temperature parameters, which are derived based on regression analysis of historical temperature sensor data. The rolling update of the calibration reference table is achieved by replacing the earliest historical version. The historical version data is stored with a snapshot of the environmental parameters at the time of version generation.

4. A health monitoring and management system based on multimodal indicators according to claim 3, characterized in that: The analysis process of the signal offset mode is as follows: The original signal of the wearable device is divided into independent segments according to the heartbeat cycle, with the R wave peak point as the cycle starting point. Each independent segment is waveform-superimposed with the corresponding periodic reference signal of the professional medical device, and the absolute timestamp of the R wave peak point is aligned during superposition. The amplitude difference area of ​​the superimposed waveform is calculated, and the amplitude difference area is the vertical integral difference of the two waveforms. The cumulative delay is calculated as the sum of the time offsets of the two waveforms at the same characteristic point. The fixed deviation component caused by the inherent characteristics of the equipment in the amplitude difference area is identified. The fixed deviation component is obtained through statistical analysis of historical data. The random fluctuation component caused by ambient temperature fluctuation is separated. The random fluctuation component is extracted using a sliding window filtering algorithm. The fixed deviation component is written into the calibration rules as the basic correction quantity, and the random fluctuation component generates a dynamic compensation term linked to the ambient temperature.

5. The health monitoring and management system based on multimodal indicators according to claim 1, characterized in that: The signal correction process of the dynamic correction unit includes: Retrieve a correction parameter combination that matches the current timestamp, ambient temperature, and device wearing status from the calibration table; perform a baseline reset on the wearable device's original signal, using the average amplitude of the reference signal over the same time period as the baseline; apply an amplitude correction factor and a delay compensation amount, with the correction factor weighted by the frequency component of the signal in each frequency band; and reconstruct the data stream using waveform splicing technology, using the R-wave apex of the reference signal as the time alignment anchor point. During the time domain alignment process, the sampling points of the wearable device signal are interpolated and resampled to make its time resolution consistent with the reference signal; in the reconstructed data stream, the difference between the QRS complex width of each heartbeat cycle and the corresponding value of the reference signal is controlled within twice the sampling interval; the corrected signal stream is transmitted to the calibration verification unit in real time, and the version identifier of the calibration reference table currently in use is attached during the transmission.

6. The health monitoring and management system based on multimodal indicators according to claim 1, characterized in that: In the calibration verification unit, the matching process of the calibration window is: When the user is at rest and the ambient temperature fluctuation is within the set range, the calibration window is automatically triggered to open. The duration of the calibration window covers at least 20 consecutive heartbeat cycles to ensure that signal characteristics under different heart rate conditions are included. The matching error is calculated using a dynamic time warping algorithm, and local path constraints are set in the algorithm to eliminate accidental offsets. The error threshold is dynamically adjusted according to the measurement accuracy of professional medical equipment, and the adjustment range is linked to the technical specifications of the equipment. Updates to the calibration rule base include replacing new entries with historical entries, and verifying the generalization capabilities of new entries within the historical dataset. Trend analysis of historical version data uses a moving average algorithm to identify the drift patterns of calibration parameters over time. The trend analysis results are used to generate optimization suggestions for the calibration table, which include pre-adjustments for the compensation coefficients.

7. The health monitoring and management system based on multimodal indicators according to claim 6, characterized in that: The dynamic adjustment process of the error threshold is as follows: Obtain the signal stability indicators of professional medical equipment within the last 24 hours, including amplitude standard deviation and delay jitter. Calculate the basic error tolerance based on the stability indicators, which is proportional to the standard deviation. Add the device usage attenuation factor, which increases linearly with the device's cumulative operating time. The final error threshold is set as the product of the basic error tolerance and the attenuation factor, and the product result is limited to the maximum allowable error range of the equipment; when the matching error exceeds the final threshold, the version rollback mechanism of the calibration rule library is triggered; the version rollback mechanism retrieves the three most recent valid versions from the archive storage area and selects the historical version closest to the current environmental parameters; after the rollback operation is completed, the calibration verification process is restarted and the current invalid rule entry is disabled.

8. The health monitoring and management system based on multimodal indicators according to claim 1, characterized in that: The historical version data specifically includes: During the daily system maintenance window, data cleaning is performed on historical versions to remove invalid parameter combination entries. Valid entries are stored according to the ambient temperature gradient, with the classification interval being the set temperature value, and overlapping buffer zones exist between adjacent classification intervals. When trend analysis calls historical versions, data from consecutive time periods under the same temperature classification is prioritized. The drift pattern of the calibration parameters is fitted to the slope of a straight line using the least squares method, and the slope value is used to predict the correction increment for the next cycle. The prediction result is converted into pre-adjustment instructions for the calibration comparison table, which are automatically loaded at the beginning of the next calibration cycle. The application of pre-adjustment instructions must go through a temporary verification process. The temporary verification uses a shortened calibration window to quickly verify the rationality of the prediction; the pre-adjustment instructions that pass the verification are merged into the formal calibration rules, and the prediction optimization mark is marked in the version identifier.

9. The health monitoring and management system based on multimodal indicators according to claim 1, characterized in that: When the wearable device's signal amplitude is detected to continuously exceed the device's measurement range, the emergency calibration mode is activated. This mode forces the use of the latest reference signal from the professional medical device, skipping the regular calibration verification process. The reference signal's capture period is extended forward to before the start of the abnormal data to ensure that normal signal characteristics are covered. When reconstructing data, a bidirectional interpolation method is used to infer the missing waveform by combining the normal data before and after the abnormal period. The data segment after interpolation and reconstruction is marked as temporary calibration data, and the temporary data is converted into official data after passing the supplementary calibration verification. The supplementary calibration verification adopts a high-density sampling mode, and the baseline data volume of high-density sampling is increased to three times that of the normal mode; the temporary data that passes the verification triggers the priority update of the calibration rule library, and the effective time period of the new rule entry is shortened to the set proportion of the original value.

Images