A non-invasive blood glucose detection method based on a wearable device

Abstract

This invention belongs to the field of blood glucose detection technology, specifically disclosing a non-invasive blood glucose detection method based on wearable devices. The method includes: matching an initial blood glucose time-series graph from a blood glucose database based on the user's initial information, and dynamically fusing it with blood glucose values ​​collected during the first usage cycle to generate a personalized blood glucose time-series graph; real-time detection of blood glucose values, influencing parameters, and heart rate signals; correcting blood glucose values ​​using a multivariate compensation model, while simultaneously calculating heart rate variability characteristics based on the heart rate signals; comparing the corrected blood glucose values ​​with the personalized blood glucose fluctuation range, and simultaneously comparing the heart rate variability characteristics with a preset deviation threshold to comprehensively determine whether the blood glucose is abnormal; if abnormal, dynamically adjusting the blood glucose detection frequency by identifying the proportion of abnormal blood glucose and the heart rate variability trend. This invention improves the accuracy of non-invasive blood glucose detection through personalized calibration, multi-parameter compensation, collaborative determination of heart rate variability, and frequency adjustment.

Description

Technical Field

[0001] This invention belongs to the field of blood glucose detection technology and relates to a non-invasive blood glucose detection method based on wearable devices. Background Technology

[0002] Non-invasive blood glucose testing technology aims to avoid the pain and infection risks associated with traditional finger-prick blood sampling, thereby improving monitoring adherence among diabetic patients. Currently, most mainstream non-invasive testing methods are based on principles such as photoplethysmography, electrochemical impedance spectroscopy, and metabolic thermal integration. These methods use wearable sensors to collect optical, electrical, or thermal signals from human tissues and then use machine learning or statistical models to infer blood glucose concentration. However, due to the influence of individual metabolic differences, skin condition, ambient temperature, and exercise interference on blood glucose levels, the accuracy of existing non-invasive testing methods still faces significant challenges.

[0003] For example, Chinese invention patent CN114397334A discloses a non-invasive blood glucose analysis system, which includes a data acquisition device, a mobile terminal, and a server. The data acquisition device, through a main control module, indicator light module, data acquisition module, Bluetooth communication module, and battery management module, performs self-testing, power management, Bluetooth connection, and collects data on the target object and the testing environment. The mobile terminal displays the PPG signal and uploads the collected data to the server. After completing the blood glucose analysis, the server sends the analyzed data back to the mobile terminal for display. This system reduces the difficulty of testing for users to a certain extent and improves the user experience.

[0004] However, the aforementioned existing technologies and most current non-invasive blood glucose testing solutions still have the following shortcomings: First, existing technologies only acquire raw data such as PPG signals through data acquisition devices, and then perform uniform blood glucose analysis by the server. They do not consider matching the initial blood glucose time series based on the user's initial information, nor do they dynamically fuse the measured blood glucose value with the initial time series within the first usage cycle to generate a personalized blood glucose fluctuation range. As a result, they cannot adapt to the individual differences of different users, and the benchmark for blood glucose judgment remains unchanged during long-term monitoring. It is difficult to track the individualized pattern of blood glucose fluctuations of users, and misjudgments or omissions are likely to occur.

[0005] Second, existing technologies rely solely on PPG signals for blood glucose analysis, without addressing the dynamic compensation of influencing parameters such as skin conductivity, acceleration modulus, ambient light intensity, and temperature on blood glucose measurement values. This fails to guarantee the accuracy of blood glucose detection. Furthermore, the lack of heart rate variability features for collaborative anomaly detection makes it impossible to adaptively adjust the detection frequency based on the proportion of abnormal blood glucose levels and heart rate variability trends.

[0006] Therefore, there is an urgent need for a non-invasive blood glucose detection method that can achieve individualized initial calibration, multi-parameter dynamic compensation, collaborative determination of heart rate variability, and adaptive frequency adjustment to improve detection accuracy. Summary of the Invention

[0007] In view of this, in order to solve the problems mentioned in the background art, a non-invasive blood glucose detection method based on wearable devices is proposed.

[0008] The objective of this invention can be achieved through the following technical solution: This invention provides a non-invasive blood glucose detection method based on wearable devices, comprising: matching the corresponding initial blood glucose time series graph from the blood glucose database based on the initial information input by the user; collecting the user's blood glucose value at a preset detection frequency within the first usage cycle; and dynamically fusing the blood glucose value with the value of the corresponding time period in the initial blood glucose time series graph to generate a personalized blood glucose time series graph for the user.

[0009] The system monitors the user's blood glucose level, influencing parameters, and heart rate signal in real time. Using a pre-trained multivariate compensation model, the blood glucose level is corrected based on the influencing parameters to obtain a corrected blood glucose level. At the same time, the system calculates heart rate variability characteristics based on heart rate signals within a preset time period.

[0010] The corrected blood glucose value is compared with the corresponding blood glucose fluctuation range in the personalized blood glucose time series graph, and the heart rate variability characteristics are compared with the preset deviation threshold to comprehensively determine whether the blood glucose is abnormal.

[0011] If blood glucose is abnormal, the system continuously collects and corrects blood glucose and heart rate signals at the current detection frequency within a preset time window to identify the proportion of abnormal blood glucose and the trend of heart rate variation, so as to adjust the blood glucose detection frequency.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention generates a personalized blood glucose fluctuation range by matching the initial blood glucose time series diagram according to the user's initial information and dynamically integrating the measured blood glucose value with the time series diagram during the first use period, so that the blood glucose judgment benchmark can adapt to the individual differences of different users, thereby improving the accuracy and individual adaptability of long-term monitoring.

[0013] (2) This invention detects the influencing parameters in real time, calculates the correction coefficients corresponding to the deviation of each parameter using a pre-trained multivariate compensation model, and adds all the correction values ​​to the original blood glucose value to obtain the corrected blood glucose value. This avoids the influence of external factors on the PPG signal fluctuation, thereby improving the accuracy of non-invasive blood glucose measurement in different usage scenarios.

[0014] (3) This invention extracts the RR interval sequence from the heart rate signal, calculates the root mean square of the difference between adjacent RR intervals using a sliding window, and compares it with the user's personal baseline to obtain the relative change rate as the heart rate variability feature value. Then, it arranges the heart rate variability time series in chronological order, providing a collaborative judgment basis independent of blood glucose value for subsequent blood glucose abnormality warning.

[0015] (4) This invention compares the corrected blood glucose value with the fluctuation range of the corresponding time period in the personalized blood glucose time series chart, and compares the absolute value of the heart rate variability feature value of the current window with the preset deviation threshold. When any condition exceeds the range, blood glucose is determined to be abnormal, thus realizing the dual verification of blood glucose value and physiological state and avoiding false alarms or omissions of a single indicator.

[0016] (5) The present invention identifies the deterioration, recovery or stabilization trend by statistically analyzing the proportion of abnormal points exceeding the blood glucose fluctuation range within a preset time window after determining blood glucose abnormality, and calculating the linear regression slope of the heart rate variability characteristic value. Then, the frequency correction coefficient is dynamically calculated based on the deviation between the abnormal proportion and the expected proportion. The correction factor is matched according to the trend type, and the current frequency is multiplied by the two and compared with the highest frequency. The smaller one is taken as the adjusted frequency. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram showing the connections between the steps of the method of the present invention.

[0019] Figure 2 This is a schematic diagram showing the connection steps of the personalized blood glucose time series diagram generation method of the present invention.

[0020] Figure 3 This is a schematic diagram showing the connection steps for adjusting the blood glucose detection frequency in this invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention improves the accuracy of non-invasive blood glucose testing through systematic personalized calibration, multi-parameter dynamic compensation, collaborative determination of heart rate variability, and adaptive frequency adjustment. Specifically, the method first matches an initial blood glucose time series graph based on the user's initial information and then merges measured blood glucose values ​​within the first usage cycle to generate a personalized blood glucose fluctuation range. Next, it detects influencing parameters such as skin conductivity, acceleration modulus, ambient light intensity, and temperature in real time, using a pre-trained multivariate compensation model to correct the original blood glucose value, while simultaneously extracting heart rate variability features from the heart rate signal. The corrected blood glucose value is then compared with the personalized blood glucose fluctuation range, and the degree of deviation from the heart rate variability features is used to comprehensively determine whether the blood glucose is abnormal. Finally, when an abnormality is determined, the percentage of abnormal blood glucose and the trend of heart rate variability within a preset time window are statistically analyzed, and the detection frequency is dynamically adjusted.

[0023] Please see Figure 1 As shown, the non-invasive blood glucose detection method based on wearable devices provided by the present invention includes the following steps S1 to S4.

[0024] S1. Generate the user's personalized blood glucose time-series chart. Please see Figure 2 As shown, this step aims to construct a personalized time-series graph reflecting the individual user's blood glucose fluctuation pattern based on the user's initial information and measured blood glucose values ​​during the first usage cycle, providing an adaptive benchmark for subsequent anomaly detection. Specifically, it includes the following sub-steps: S1-1. Obtain the initial lower limit and initial upper limit of the initial blood glucose fluctuation range within each time period from the initial blood glucose time series graph. The initial blood glucose time series graph is obtained by matching the user-inputted initial information such as age, weight, and diabetes type from a preset blood glucose database.

[0025] S1-2. Obtain all measured blood glucose values ​​for each time period during the first usage cycle and count the number of measurements for each time period.

[0026] S1-3. If the number of actual measurements is greater than 0, calculate the minimum and maximum values ​​of the actual measurements for that period.

[0027] S1-4. Multiply the actual number of measurements by the actual minimum value, then add this product to the initial lower limit to obtain the lower limit sum. Divide this lower limit sum by the actual number of measurements plus 1 to obtain the personalized lower limit. The personalized lower limit... The calculation formula is: In the formula The actual number of measurements. This is the measured minimum value. This is the initial lower bound.

[0028] S1-5. Multiply the actual number of measurements by the actual maximum value, then add this to the initial upper limit to obtain the upper limit sum. Divide this upper limit sum by the actual number of measurements plus 1 to obtain the personalized upper limit. The personalized lower limit and personalized upper limit then constitute the personalized fluctuation range for this period. The personalized upper limit... The calculation formula is: In the formula This is the measured maximum value. This is the initial upper limit.

[0029] S1-6. If the number of actual measurements is 0, obtain the personalized fluctuation range of adjacent time periods and calculate the personalized fluctuation range of the time period.

[0030] Furthermore, the calculation of the personalized fluctuation range for this period includes: S1-6-1. Determine all adjacent time periods of this time period, that is, the actual time periods that exist in the previous time period and the next time period, and then count the number of adjacent time periods.

[0031] S1-6-2. If the number of adjacent time periods is 2, i.e. non-boundary time periods, then calculate the lower average and upper average of the personalized fluctuation range of the adjacent time periods respectively, and use the average as the personalized lower limit and personalized upper limit of the time period.

[0032] S1-6-3. If the number of adjacent time periods is 1, i.e., the boundary time period, then the personalized fluctuation range of the adjacent time periods shall be taken as the personalized fluctuation range of the current time period.

[0033] S1-6-4. If the number of adjacent time periods is 0, the initial blood glucose fluctuation range is taken as the personalized fluctuation range for that time period. Specifically, the personalized fluctuation ranges for all time periods with a measured number greater than 0 are calculated first, and then the time periods with a measured number of 0 are processed.

[0034] S1-7. Combine the personalized fluctuation ranges of all time periods to generate a personalized blood glucose time series chart for the user.

[0035] Furthermore, to maintain the long-term adaptability of the blood glucose fluctuation range to the user's current physiological state, the personalized blood glucose time series chart can be periodically updated based on the measured blood glucose values ​​continuously generated by the user during use. Specifically, every preset update cycle (e.g., 7 days or 30 days), all measured blood glucose values ​​collected within the update cycle are re-fused according to steps S1-1 to S1-7 to generate an updated personalized blood glucose time series chart, which replaces the original time series chart for subsequent anomaly detection. If the number of measured blood glucose values ​​collected in the current update cycle is less than the preset minimum sample size (e.g., 80% of the total amount to be collected), the update will not be performed temporarily, the original version will be used, and an update attempt will be made in the next update cycle.

[0036] S2. Real-time monitoring of blood glucose levels, influencing parameters, and heart rate signals to obtain corrected blood glucose levels and heart rate variability characteristics. This step involves two parallel processing steps: first, using a multivariate compensation model to correct the original blood glucose value and eliminate environmental and exercise interference; second, extracting the heart rate variability feature value sequence from the heart rate signal. Specifically, it consists of the following sub-steps.

[0037] S2-1. Pre-train a multivariate compensation model (this can be done offline and called during real-time detection). S2-1-1. Set standard conditions in a laboratory environment. The standard conditions include: the subject is in a seated state, without movement, with dry skin, the ambient temperature is controlled between 22°C and 26°C, and the ambient light intensity is less than 100 lux. Collect the raw blood glucose values ​​and various influencing parameters output by the wearable device from multiple subjects. The influencing parameters include skin conductivity, acceleration modulus, ambient light intensity, and temperature. Then, the average value of each influencing parameter under the standard conditions is used as the benchmark value of the corresponding influencing parameter.

[0038] S2-1-2. Modify each individual influencing parameter (e.g., gradually increase the ambient temperature from a baseline of 22℃ to 35℃), while simultaneously controlling the deviation of other parameters from their respective baseline values ​​to not exceed preset stability conditions. When the modified influencing parameters are within the preset stability conditions and the duration reaches a preset stability period, simultaneously record the original blood glucose value from the wearable device and the reference blood glucose value measured via finger-prick blood sampling. The preset stability condition is a deviation not exceeding ±5% of the baseline value, and the preset stability period is 30 seconds.

[0039] S2-1-3. Subtract the baseline value from the measured value of each influencing parameter to obtain the deviation. Then, divide the range of deviation values ​​into multiple continuous and non-overlapping deviation intervals. Collect all samples within each deviation interval, calculate the difference between the reference blood glucose value and the original blood glucose value for each sample, and then calculate the average of all differences within each deviation interval as the total correction amount for the corresponding deviation interval. The deviation intervals can be divided using equal intervals or based on quantiles of the sample distribution, with 5 to 10 deviation intervals.

[0040] S2-1-4. Calculate the midpoint value of each deviation interval. This midpoint value is the average of the lower and upper limits of the interval. Divide the total correction amount within each deviation interval by the midpoint value of that deviation interval to obtain the correction coefficient corresponding to that deviation interval. If the midpoint value of a deviation interval is zero, then theoretically the total correction amount for that interval should be zero, and the correction coefficient is set to zero in this case.

[0041] S2-1-5. Store the correction coefficients of each influencing parameter in each deviation range as a multivariate compensation model.

[0042] S2-2, Real-time correction of blood glucose levels S2-2-1. Calculate the deviation of each influencing parameter from its benchmark value. The deviation is the measured value minus the benchmark value, and can be positive or negative.

[0043] S2-2-2. Determine the deviation range based on the values ​​of the deviations of each influencing parameter, and query the correction coefficient corresponding to the deviation range from the multivariate compensation model. The correction coefficient is signed and indicates the direction and magnitude of the compensation required per unit deviation.

[0044] S2-2-3. Multiply the deviation of the influencing parameter by the corresponding correction coefficient to obtain the correction amount of each influencing parameter.

[0045] S2-2-4. Add the correction values ​​of all influencing parameters to the blood glucose value to obtain the corrected blood glucose value.

[0046] S2-3, Calculate heart rate variability characteristics Heart rate variability reflects the autonomic nervous system's ability to regulate heart rhythm, and its changes are correlated with abnormal blood glucose events. To obtain heart rate variability characteristics, the timing of each heartbeat needs to be extracted from the heart rate signal. In this embodiment, the wearable device collects pulse wave signals through a PPG sensor, identifies the peak positions using a peak detection algorithm, and the time interval between adjacent peaks is the RR interval, thus forming an RR interval sequence.

[0047] S2-3-1. Extract the interval time of continuous heartbeats from the heart rate signal to obtain the RR interval sequence.

[0048] S2-3-2. Using a sliding time window, calculate the square of the difference between all adjacent RR intervals within the window, sum all the square values, divide by the number of adjacent intervals, and then take the square root to obtain the root mean square of the difference between adjacent RR intervals in each window.

[0049] In one specific embodiment, let there be a total of within the sliding time window The RR interval is denoted as Then the difference between adjacent RR intervals is: - , Then the root mean square The calculation formula is: .

[0050] S2-3-3. Compare the root mean square (RMS) with the user's baseline value, calculate the relative rate of change, and use it as the heart rate variability characteristic value for each window. The user's baseline value is acquired and stored in the wearable device in the following ways: Firstly, the RMSSD is automatically calculated and updated using stable heart rate signals during the user's sleep period; if no sleep data is available, the user is prompted to actively collect signals for 5 to 10 minutes in a resting state before calculating the RMSSD; if a personal baseline value cannot be obtained, a healthy person's RMSSD reference value (e.g., 30ms) can be used as a temporary baseline and continuously updated during subsequent use.

[0051] S2-3-4. Arrange the relative rates of change of each window in chronological order to form a time series of heart rate variability.

[0052] S3. Comprehensive assessment of whether blood sugar is abnormal. This step will correct the blood glucose value and compare it with the personalized blood glucose fluctuation range. At the same time, it will compare the heart rate variability feature value of the current window with the preset deviation threshold and use OR logic to make a comprehensive judgment.

[0053] S3-1. Compare the corrected blood glucose value with the blood glucose fluctuation range of the corresponding time period in the user's personalized blood glucose time series graph, and compare the heart rate variability feature value of the current window with the preset deviation threshold.

[0054] The preset deviation threshold is used to determine whether the heart rate variability characteristic value of the current window deviates from the individual's baseline. It is obtained by using the historical characteristic value distribution of the user at rest for multiple consecutive days at the same time, and taking the 95% confidence interval boundary under a normal distribution as the deviation threshold. Upon first use, a group experience value is used as the deviation threshold, for example, set as a relative change rate of ±20%, i.e., an absolute value of 0.2. This experience value is obtained by statistically analyzing the heart rate variability characteristic distribution of multiple healthy adult subjects at rest, and its 95% confidence interval boundary is approximately ±20%.

[0055] S3-2. When the corrected blood glucose value exceeds the blood glucose fluctuation range of the corresponding time period in the personalized blood glucose time series chart, or the absolute value of the heart rate variability feature is greater than the preset deviation threshold, it is determined to be abnormal blood glucose; otherwise, it is determined to be normal blood glucose.

[0056] S4. If blood sugar is abnormal, adjust the frequency of blood sugar testing. Please see Figure 3 As shown, when S3 determines that blood glucose is abnormal, this step is executed. By analyzing the proportion of abnormal blood glucose and the trend of heart rate variation within the abnormal window, the frequency of subsequent tests is dynamically adjusted.

[0057] S4-1, Identifying the percentage of individuals with abnormal blood glucose levels S4-1-1. Within the preset time window, continuously collect the corrected blood glucose values ​​at each detection time point according to the current detection frequency.

[0058] S4-1-2. Compare each corrected blood glucose value with the corresponding blood glucose fluctuation range in the personalized blood glucose time series chart, and count the number of abnormal detection time points that exceed the range.

[0059] S4-1-3. The ratio of the number of abnormal detection time points to the total number of detection points within the time window is taken as the percentage of abnormal blood glucose.

[0060] S4-2, Identifying Heart Rate Variation Trends The preset normal slope range is used to determine whether the heart rate variability trend is in a stable state. Its lower limit is negative and its upper limit is positive. When used for the first time, empirical values ​​based on clinical population data are used, for example, set to [-0.05, 0.05]. During use, the user can select the median ± 1 standard deviation of the heart rate variability time series slope distribution of their own historical stable period (such as the period of no abnormal blood glucose in the past 7 days) as the personalized normal slope range, and update it regularly.

[0061] S4-2-1. Calculate the linear regression slope of the heart rate variability time series. If the slope exceeds the preset normal slope range, a positive slope indicates a worsening trend in heart rate variability, while a negative slope indicates a recovery trend in heart rate variability.

[0062] S4-2-2. If the slope is within the preset normal slope range, it is determined to be a stable trend. If the slope is equal to the upper or lower limit of the preset normal slope range, it is also determined to be a stable trend.

[0063] S4-3. Calculate the adjusted detection frequency. S4-3-1. Obtain the percentage of abnormal blood glucose levels within a preset time window, and dynamically calculate the frequency correction coefficient based on the deviation between this percentage and the expected percentage of abnormalities.

[0064] The expected abnormality percentage refers to the proportion of time within a historical time window that is expected to result in a blood glucose abnormality, based on the user's historical health status and blood glucose control goals. This percentage is a baseline target value used to measure whether current blood glucose fluctuations are within the normal range. It is set based on the median of the actual abnormality percentages during a stable period (e.g., 7 days) of the user's past data.

[0065] Frequency correction factor The formula used to adjust the subsequent data acquisition frequency in real time employs a proportional control method: In the formula This represents the percentage of time with abnormal blood glucose levels within a preset time window. The expected percentage of abnormalities, A fixed gain coefficient is used to control the adjustment level. The desired abnormal percentage... It is generally set to 0.1 (i.e., considered abnormal within 10% of the time window), and this value is based on the frequency of blood glucose fluctuations in the normal population according to clinical statistics; fixed gain coefficient. The value ranges from 0.5 to 2.0. The specific value can be pre-calibrated based on the user's historical blood glucose fluctuation characteristics, or dynamically adjusted during system operation using adaptive control methods (such as fuzzy logic or gain scheduling). For example, when the system detects severe fluctuations in the user's blood glucose, a larger value, such as 1.5, is used to speed up the frequency adjustment response; when the user is asleep or at rest and their blood glucose is relatively stable, a smaller value, such as 0.8, is used to avoid frequent frequency fluctuations.

[0066] when hour, Without changing the fundamental frequency or sensitivity. At that time, the deviation is positive. This indicates that abnormal blood glucose levels are occurring too frequently, requiring a higher sampling frequency to capture more details. When the deviation is negative, This indicates that blood glucose levels are abnormally low. It is advisable to appropriately reduce the sampling frequency or extend the window to reduce the computational burden.

[0067] S4-3-2. Obtain the heart rate variability trend within a preset time window, whereby the heart rate variability trend is categorized into deterioration trend, recovery trend, and stable trend. Match the corresponding correction factor based on the heart rate variability trend type.

[0068] It should be added that the correction factor is used to dynamically adjust the monitoring sensitivity of abnormal blood glucose based on heart rate variability trends (such as sampling frequency and judgment threshold). It is obtained using a two-step method: a base empirical value plus user-provided historical corrections. The first step is to set basic empirical values: Based on clinical population data, predefine the basic correction factors for each trend type: the basic correction factor for a deteriorating trend is set to be greater than 1, for example, 1.2; the basic correction factor for a recovery trend is set to be less than 1, for example, 0.8; and the basic correction factor for a stable trend is set to 1.

[0069] The second step involves personalized adjustments using the user's historical data: collecting various trend window data from the user over a past period (e.g., 30 days), along with the corresponding blood glucose anomaly detection performance (e.g., the weighted sum of false positive and false negative rates). The moving average that minimizes the detection cost under the same historical trend is then calculated and used as the correction factor for each heart rate variability trend type.

[0070] S4-3-3. Multiply the current detection frequency with the frequency correction coefficient and the correction factor to obtain the corrected detection frequency.

[0071] S4-3-4. Compare the adjusted detection frequency with the preset maximum detection frequency, and take the smaller one as the adjusted blood glucose detection frequency. In addition, the system also has a preset minimum detection frequency. When the adjusted detection frequency is lower than the minimum frequency, the minimum frequency is taken as the final detection frequency to ensure that abnormal blood glucose events can be captured in a timely manner.

[0072] The preset maximum detection frequency refers to the upper limit of the maximum frequency of blood glucose detection allowed by the system. This is used to prevent excessive power consumption, data redundancy, or unnecessary interference to the user due to over-detection. It is obtained by calculating the highest stable detection frequency that can be supported based on the wearable device's sensor maximum sampling rate, processor processing power, memory capacity, and battery life requirements. For example, if a photoelectric sensor supports a maximum blood glucose detection every 5 minutes, then the maximum frequency is 0.2 times / minute.

[0073] Through steps S1 to S4 described above, this invention achieves a complete closed loop from individualized initial calibration, multi-parameter dynamic compensation, collaborative determination of heart rate variability, to adaptive frequency adjustment driven by anomalies. Compared with existing technologies, this method significantly improves the accuracy, timely warning, and energy efficiency of non-invasive blood glucose detection.

[0074] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0075] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0076] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0077] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0078] Finally, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A non-invasive blood glucose detection method based on wearable devices, characterized in that: The method includes: Based on the initial information input by the user, the system matches the corresponding initial blood glucose time series graph from the blood glucose database. During the first usage cycle, the system collects the user's blood glucose value at a preset detection frequency and dynamically merges the blood glucose value with the value of the corresponding time period in the initial blood glucose time series graph to generate the user's personalized blood glucose time series graph. The system monitors the user's blood glucose level, influencing parameters, and heart rate signal in real time. Using a pre-trained multivariate compensation model, the blood glucose level is corrected according to the influencing parameters to obtain the corrected blood glucose level. At the same time, the system calculates the heart rate variability characteristics based on the heart rate signal within a preset time period. The corrected blood glucose value is compared with the corresponding blood glucose fluctuation range in the personalized blood glucose time series graph, and the heart rate variability characteristics are compared with the preset deviation threshold to comprehensively determine whether the blood glucose is abnormal. If blood glucose is abnormal, the system continuously collects and corrects blood glucose and heart rate signals at the current detection frequency within a preset time window to identify the proportion of abnormal blood glucose and the trend of heart rate variation, so as to adjust the blood glucose detection frequency.

2. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The process of generating a personalized blood glucose time-series graph for the user includes: Obtain the initial lower limit and initial upper limit of the initial blood glucose fluctuation range for each time period from the initial blood glucose time series graph; Obtain all measured blood glucose values ​​of the user at each time period during the first usage cycle, and count the number of measurements at each time period; If the number of actual measurements is greater than 0, calculate the minimum and maximum values ​​of the actual measurements for that period. Multiply the actual number of measurements by the actual minimum value and add it to the initial lower limit to obtain the lower limit sum. Then divide the lower limit sum by the actual number of measurements plus 1 to obtain the personalized lower limit. Multiply the number of measurements by the maximum value and add it to the initial upper limit to get the upper limit sum. Then divide the upper limit sum by the number of measurements plus 1 to get the personalized upper limit. The personalized lower limit and personalized upper limit constitute the personalized fluctuation range for this period. If the number of actual measurements is 0, then obtain the personalized fluctuation range of the adjacent time periods and calculate the personalized fluctuation range of the time period. The personalized fluctuation ranges of all time periods are combined to generate a personalized blood glucose time series chart for each user.

3. The non-invasive blood glucose detection method based on a wearable device according to claim 2, characterized in that: The calculation of the personalized fluctuation range for this period includes: Identify all adjacent time periods of this time period, and then count the number of adjacent time periods; If the number of adjacent time periods is 2, then calculate the lower average and upper average of the personalized fluctuation range of the adjacent time periods respectively, and use the average as the personalized lower limit and personalized upper limit of the time period. If the number of adjacent time periods is 1, then the personalized fluctuation range of the adjacent time periods will be used as the personalized fluctuation range of the current time period. If the number of adjacent time periods is 0, then the initial blood glucose fluctuation range will be used as the personalized fluctuation range for that time period.

4. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The acquisition of the pre-trained multivariate compensation model includes: In a laboratory setting, standard conditions were set, and raw blood glucose values ​​and various influencing parameters were collected from multiple subjects using wearable devices. The average value of each influencing parameter under standard conditions was then used as the baseline value for the corresponding influencing parameter. Each single influencing parameter is changed separately, while controlling the deviation of other parameters from their respective benchmark values ​​to not exceed the preset stability conditions. When the changed influencing parameter is within the preset stability conditions and the duration reaches the preset stability duration, the original blood glucose value of the wearable device and the reference blood glucose value measured by finger prick blood sampling are recorded simultaneously. The deviation is obtained by subtracting the baseline value from the measured value of each influencing parameter. The range of deviation is then divided into multiple continuous and non-overlapping deviation intervals. All samples in each deviation interval are collected, and the difference between the reference blood glucose value and the original blood glucose value of each sample is calculated. The average value of all differences in each deviation interval is then calculated as the total correction amount for the corresponding deviation interval. Calculate the midpoint value of each deviation interval. This midpoint value is the average of the lower and upper limits of the interval. Divide the total correction amount in each deviation interval by the midpoint value of that deviation interval to obtain the correction coefficient corresponding to that deviation interval. The correction coefficients of each influencing parameter within each deviation range are stored as a multivariate compensation model.

5. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The calculation of the corrected blood glucose value includes: Calculate the deviation of each influencing parameter from its baseline value; The deviation range is determined based on the value of the deviation of each influencing parameter, and the correction coefficient corresponding to the deviation range is queried from the multivariate compensation model. Multiply the deviation of the influencing parameter by the corresponding correction factor to obtain the correction amount of each influencing parameter; Add the correction values ​​for all influencing parameters to the blood glucose value to obtain the corrected blood glucose value.

6. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The calculated heart rate variability features include: The interval between consecutive heartbeats is extracted from the heart rate signal to obtain the RR interval sequence; Using a sliding time window, the square of the difference between all adjacent RR intervals within the window is calculated. The sum of all squares is then divided by the number of adjacent intervals, and the square root is taken to obtain the root mean square of the difference between adjacent RR intervals in each window. The root mean square value is compared with the user's baseline value to calculate the relative rate of change, which is then used as the heart rate variability feature value for each window. The relative rates of change of each window are arranged in chronological order to form a time series of heart rate variability.

7. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The comprehensive determination of whether blood glucose is abnormal includes: The corrected blood glucose value is compared with the blood glucose fluctuation range of the corresponding time period in the user's personalized blood glucose time series graph, and the heart rate variability feature value of the current window is compared with the preset deviation threshold. When the corrected blood glucose value exceeds the blood glucose fluctuation range of the corresponding time period in the personalized blood glucose time series chart, or when the absolute value of the heart rate variability feature is greater than the preset deviation threshold, it is determined to be abnormal blood glucose; otherwise, it is determined to be normal blood glucose.

8. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The percentage of abnormal blood glucose levels identified includes: Within the preset time window, corrected blood glucose values ​​are continuously collected at each testing time point according to the current testing frequency; Compare each corrected blood glucose value with the corresponding blood glucose fluctuation range in the personalized blood glucose time series chart, and count the number of abnormal detection time points that exceed the range; The ratio of the number of abnormal detection time points to the total number of detection points within the time window is used as the percentage of abnormal blood glucose levels.

9. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The identification of heart rate variability trends includes: Calculate the linear regression slope of the heart rate variability time series. If the slope exceeds the preset normal slope range, a positive slope indicates a worsening trend of heart rate variability, while a negative slope indicates a recovery trend of heart rate variability. If the slope is within the preset normal slope range, it is determined to be a stable trend.

10. The non-invasive blood glucose detection method based on a wearable device according to claim 1, characterized in that: The adjustment of blood glucose detection frequency includes: Obtain the percentage of abnormal blood glucose levels within a preset time window, and dynamically calculate the frequency correction coefficient based on the deviation between this percentage and the expected percentage of abnormalities. Obtain the heart rate variability trend within a preset time window, and match the corresponding correction factor according to the heart rate variability trend type; The corrected detection frequency is obtained by multiplying the current detection frequency by the frequency correction coefficient and the correction factor. The corrected detection frequency is compared with the preset maximum detection frequency, and the smaller one is taken as the adjusted blood glucose detection frequency.

Images