A non-invasive-minimally invasive blood glucose prediction system, method and storage medium
By using a non-invasive and minimally invasive combined blood glucose prediction system, which deeply integrates blood glucose parameters and physiological parameters through GRU networks and fully connected networks, the system solves the problem of insufficient accuracy in blood glucose monitoring in existing technologies, and achieves accurate prediction of blood glucose changes, making it suitable for long-term management of diabetic patients.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUN YAT SEN UNIV
- Filing Date
- 2026-01-26
- Publication Date
- 2026-06-02
AI Technical Summary
Existing non-invasive blood glucose monitoring technologies suffer from insufficient accuracy, and conventional AI models lack effective integration and in-depth analysis of historical blood glucose data, failing to meet clinical needs for precise blood glucose monitoring.
Blood glucose and physiological parameters are collected using a combination of non-invasive and minimally invasive methods. These parameters are then deeply integrated through a blood glucose detection model. A GRU network is used to extract historical multi-parameter features and blood glucose trend features, which are then combined with a fully connected network to generate predicted blood glucose values.
It enables accurate prediction of blood glucose changes, improves the accuracy and reliability of monitoring, and provides a reliable long-term management tool for diabetic patients.
Smart Images
Figure CN122136012A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of blood glucose detection technology, and in particular to a non-invasive / minimally invasive blood glucose prediction system, method and storage medium. Background Technology
[0002] There are three main types of continuous glucose monitoring (CGM) technologies: invasive, minimally invasive, and non-invasive. Invasive CGM relies on methods such as finger-prick blood sampling. While it can obtain accurate blood glucose values, it suffers from drawbacks such as invasiveness, cumbersome operation, susceptibility to infection, and inability to provide continuous monitoring, severely impacting patient compliance. Minimally invasive CGM devices can achieve short-term continuous monitoring, but electrode enzyme degradation issues typically limit their lifespan to no more than 15 days, making it difficult to meet long-term monitoring needs. Non-invasive monitoring technology has become a research hotspot due to its advantages of being non-invasive and convenient. It mainly uses physical methods such as spectroscopy (near-infrared, mid-infrared, and Raman spectroscopy) and metabolic heat compensation, combined with artificial intelligence (AI) models to deduce blood glucose concentration. Related models encompass traditional machine learning algorithms such as multinomial regression and support vector machines, as well as deep learning algorithms such as multi-layer perceptrons, recurrent neural networks, gated recurrent units, and long short-term memory networks.
[0003] With the deepening application of AI technology in the medical field, non-invasive blood glucose monitoring has made some progress in model fitting and monitoring by collecting human physiological data. However, its clinical application is still limited by the accuracy and practicality of monitoring. Among the existing non-invasive continuous glucose monitoring technologies, devices based on electrochemical methods have limited correlation between the glucose concentration in tears and sweat and the actual blood glucose level, making it difficult to meet clinical accuracy requirements. Non-invasive technologies that use principles such as spectroscopy (e.g., near-infrared or mid-infrared spectroscopy), Raman spectroscopy, or metabolic heat compensation to infer blood glucose concentration are still in the exploration and optimization stage. At the same time, when performing non-invasive blood glucose value derivation, conventional AI models often only use the non-invasive physiological parameters collected at the current moment (such as skin temperature, ambient temperature and humidity related to metabolic heat, and heart rate, blood oxygen saturation, blood flow velocity, etc. related to cardiovascular diseases) as core input data, lacking effective integration and in-depth mining of historical blood glucose-related data, and ignoring the key trend information contained in historical blood glucose data. Summary of the Invention
[0004] The main objective of this application is to propose a non-invasive / minimally invasive blood glucose prediction system, method, and storage medium to solve the aforementioned problems.
[0005] To achieve the above objectives, one aspect of this application proposes a non-invasive / minimally invasive blood glucose prediction system, the system comprising:
[0006] The data acquisition module is used to collect blood glucose and physiological parameters through non-invasive and minimally invasive methods and input them into the blood glucose detection model; The blood glucose prediction module is used to output a predicted blood glucose value through the blood glucose detection model.
[0007] In some embodiments, the data acquisition module includes: Microneedle electrode unit is used to collect blood glucose parameters in tissue fluid; The non-invasive blood glucose sensing unit is used to collect physiological parameters through a sensor array, including historical physiological parameters and current physiological parameters.
[0008] In some embodiments, the blood glucose prediction module includes A first processing network is used to extract time-series features related to the current blood glucose value from the historical physiological parameters and generate historical multi-parameter features. The second processing network is used to analyze the historical variation patterns of the blood glucose parameters and generate blood glucose trend characteristics. A fusion network is used to fuse the historical multi-parameter features, blood glucose trend features, and current physiological parameters to generate a fusion feature vector, and then transform the fusion feature vector into a blood glucose prediction value through a multi-layer fully connected network.
[0009] In some embodiments, the converged network includes several layers of fully connected networks, the fully connected networks including: The first ReLU activation function is expressed as follows: ; The second ReLU activation function is expressed as follows: ; The blood glucose output function, the expression of which is: ; in , For the output of a fully connected network, Here, ReLU is the predicted blood glucose value, and ReLU() is the activation function. , as well as For the weight matrix of a fully connected network, , as well as This is the bias vector corresponding to the fully connected network. To fuse feature vectors.
[0010] In some embodiments, the data acquisition module further includes: The MCU module is used to convert the signals collected by the microneedle electrode unit and the non-invasive blood glucose sensor unit into blood glucose parameters and physiological parameters, respectively. Data storage module, used to store the blood glucose parameters and physiological parameters; A wireless transmission module is used to transmit the blood glucose parameters and physiological parameters to the blood glucose detection model; The power supply module is used to provide power to each module.
[0011] In some embodiments, the sensor group includes a temperature sensor, an infrared radiation sensor, and a photoelectric sensor.
[0012] In some embodiments, the step of outputting a predicted blood glucose value through the blood glucose detection model specifically includes: Extract time-series features related to current blood glucose levels from historical physiological parameters to generate historical multi-parameter features; Analyze the historical patterns of blood glucose parameters to generate blood glucose trend characteristics; The historical multi-parameter features, blood glucose trend features, and current physiological parameters are fused to generate a fused feature vector, which is then transformed into a blood glucose prediction value through a multi-layer fully connected network.
[0013] In some embodiments, the analysis of historical changes in blood glucose parameters and the generation of blood glucose trend features specifically include: Input the blood glucose parameters from the previous period into the sliding window, and calculate the overall trend characteristics of all blood glucose parameters within the sliding window; Input the blood glucose parameters of the previous period into the sliding window, divide the sliding window into several sub-windows, calculate the average blood glucose concentration of each sub-window, and calculate the average fluctuation characteristics based on the average blood glucose concentration of all sub-windows. The overall trend features and the average trend features are weighted and averaged to generate blood glucose trend features.
[0014] To achieve the above objectives, another aspect of this application proposes a non-invasive / minimally invasive blood glucose prediction method, the method comprising: Blood glucose and physiological parameters are collected through non-invasive and minimally invasive methods and then input into the blood glucose detection model; The blood glucose detection model outputs a predicted blood glucose value.
[0015] To achieve the above objectives, another aspect of the embodiments of this application proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.
[0016] The embodiments of this application include at least the following beneficial effects: This application discloses a non-invasive / minimally invasive blood glucose prediction system, method, and storage medium, which solves the limitation of conventional AI models that rely on only a single real-time data by deeply integrating blood glucose parameters and physiological parameters. Physiological parameters reflect the impact of immediate physiological state on blood glucose; blood glucose parameters reflect the current trend of blood glucose changes. The AI model establishes a correlation mechanism for blood glucose changes, which facilitates patient self-management and remote monitoring by medical staff, providing a reliable guarantee for the long-term management of diabetic patients. Attached Figure Description
[0017] Figure 1 This is a block diagram of a blood glucose prediction system provided in an embodiment of this application; Figure 2 The data acquisition module circuit diagram and working logic diagram provided in the embodiments of this application; Figure 3 A schematic diagram of the detection results of the metabolic heat and pulse wave sensors provided in the embodiments of this application; Figure 4 This is a schematic diagram of the cardiovascular sensor detection results provided in an embodiment of this application; Figure 5 This is a schematic diagram illustrating the calculation principle of blood glucose trend features provided in the embodiments of this application. Figure 6 This is a diagram illustrating the composition of the blood glucose prediction module provided in an embodiment of this application. Figure 7 This is a flowchart illustrating the process of generating predicted blood glucose values as provided in an embodiment of this application. Figure 8 A distribution map of blood glucose levels for diabetic patients and healthy subjects provided in an embodiment of this application; Figure 9 Blood glucose trend distribution graphs for diabetic patients and healthy subjects provided in the embodiments of this application; Figure 10 The blood glucose data and blood glucose trend distribution map of diabetic patients and healthy subjects provided in the embodiments of this application. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit it. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with those of this application; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this application as detailed in the appended claims.
[0019] It is understood that the terms “first,” “second,” etc., used in this application may be used herein to describe various concepts, but unless otherwise stated, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to a determination” as used herein may be interpreted as “when…” or “when…” or “in response to a determination.”
[0020] As used in this application, the terms "at least one", "multiple", "each", "any", etc., "at least one" includes one, two or more, "multiple" includes two or more, "each" refers to each of the corresponding multiples, and "any" refers to any one of the multiples.
[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.
[0022] Before providing a detailed description of the embodiments of this application, some of the nouns and terms involved in the embodiments of this application will be explained first. The nouns and terms involved in the embodiments of this application are subject to the following interpretations.
[0023] FCNN, an extended architecture of convolutional neural networks (CNN), is characterized by containing only local operations such as convolutional layers and pooling layers in the network, completely removing fully connected layers.
[0024] GRU is an efficient variant of recurrent neural networks (RNNs) designed to process sequential data (such as time series, text, and speech). It controls the flow of information through a gating structure (update gate and reset gate) to solve the long-range dependency problem of traditional RNNs (i.e., the model has difficulty remembering information from distant time steps).
[0025] THD is used to measure the linearity and fidelity of a blood glucose sensing system (or sensing unit) in the process of converting biochemical signals into electrical signals.
[0026] MARD is a statistical indicator used to quantify the overall deviation between the measurement value of a blood glucose monitoring device (such as CGM) and the reference value of venous blood or a high-precision blood glucose meter. The lower the MARD value, the higher the accuracy of the device.
[0027] BG stands for blood glucose level.
[0028] UART is an asynchronous, serial, full-duplex communication protocol and hardware interface that specifies how data can be transmitted bidirectionally bit by bit through a single data line without a unified clock signal.
[0029] Blood glucose, as a core physiological indicator of dynamic changes in the human body, forms a complex dynamic relationship with an individual's physiological state. It directly depends on immediate physiological fluctuations, including the combined influence of current metabolic levels, blood circulation, and tissue metabolic activity; it is also deeply constrained by the cumulative effect of historical blood glucose levels and fluctuation patterns, such as the baseline fasting blood glucose, the peak and decline rate of postprandial blood glucose, and the periodic characteristics of blood glucose fluctuations at different times. These historical trend information collectively constitute the "cumulative background" of blood glucose changes. Conventional AI models only accept immediate data input of current physiological state fluctuations, capturing only the instantaneous impact of the current physiological state on blood glucose, unable to trace the historical context of blood glucose changes. Consequently, the model's derivation of blood glucose values lacks historical data support and trend logic, significantly reducing accuracy and failing to meet the core clinical need for precise blood glucose monitoring.
[0030] In view of this, this application provides a non-invasive / minimally invasive blood glucose prediction system, method, and storage medium.
[0031] See Figure 1 As shown, to achieve the above objectives, this application proposes a non-invasive / minimally invasive blood glucose prediction system, the system comprising: The data acquisition module is used to collect blood glucose and physiological parameters through non-invasive and minimally invasive methods and input them into the blood glucose detection model; The blood glucose prediction module is used to output a predicted blood glucose value through the blood glucose detection model.
[0032] See Figure 2As shown, in some embodiments, the data acquisition module is an integrated wearable structure, mainly including a short-term detachable microneedle electrode unit and a non-invasive blood glucose sensing unit. The microneedle electrode unit is a microneedle electrode prepared by screen printing. The electrode is based on a biocompatible polymer material, which has both flexibility and mechanical stability. It can penetrate the stratum corneum of the skin minimally invasively and make precise contact with the tissue fluid of the dermis, while avoiding damage to subcutaneous capillaries.
[0033] The microneedle electrode is based on the second-generation glucose detection principle. It is worn on the arm for a short period of time (about 2 days) to collect the glucose concentration of tissue fluid as an accurate reference blood glucose value for blood glucose detection model calibration and historical trend feature initialization. It can be detached after calibration.
[0034] During this period, the data acquisition module continuously collects tissue fluid glucose concentration data. On the one hand, this serves as a reference blood glucose value for parameter calibration of the non-invasive blood glucose detection model, correcting systematic errors in non-invasive detection methods such as photoplethysmography (PPG). On the other hand, the blood glucose fluctuation curves obtained through continuous monitoring can be used to initialize the historical trend features of the model, providing baseline data support for subsequent blood glucose trend prediction. After the calibration and feature initialization process is completed, the module can be easily disassembled, and the single-use design effectively avoids the risk of cross-infection, meeting the convenience and safety requirements of home use.
[0035] The data acquisition module also includes: The MCU module is used to convert the signals collected by the microneedle electrode unit and the non-invasive blood glucose sensor unit into blood glucose parameters and physiological parameters, respectively. The data storage module is used to store blood glucose parameters and physiological parameters; A wireless transmission module is used to transmit blood glucose parameters and physiological parameters to the blood glucose detection model; The power supply module is used to provide power to each module.
[0036] The non-invasive blood glucose sensing unit is based on metabolic thermal integration and photoelectric sensing technology, integrating multi-dimensional sensors to continuously collect current physiological parameters. The metabolic thermal sensor includes a surface temperature sensor, an integrated radiation temperature sensor, an ambient temperature sensor, and an infrared radiation sensor, used to capture parameters such as skin surface temperature, skin radiation temperature, and ambient temperature and humidity in real time. The photoelectric sensor uses two photoplethysmography (PPG) sensors that can emit green light (530nm), red light (660nm), and infrared light (880nm) to collect cardiovascular parameters such as heart rate, blood oxygen saturation (SaO2), pulse time difference, blood flow velocity, systolic blood pressure, and diastolic blood pressure.
[0037] See Figure 3As shown, THD was continuously measured in a subject, with THD-related parameters recorded intermittently every 10 minutes during both resting and active states. PPG#1 and PPG#2 sensors were placed at the finger-wrist, wrist-wrist, finger-finger, and finger-earlobe sites, respectively, to acquire dual pulse wave signals.
[0038] See Figure 4 As shown, a metabolic thermal sensor and a photoplethysmography (PPG) sensor were fixed to the finger (PPG#1) and wrist (PPG#2) for a relatively long period (approximately 3 hours) to monitor cardiovascular parameters. Based on the dual pulse wave signals, the subject's cardiovascular parameters, including heart rate, oxygen saturation (SaO2), whole blood clotting time, systolic blood pressure, diastolic blood pressure, and blood flow velocity, were deduced.
[0039] Specifically, the blood glucose detection model focuses on the deep integration of historical multi-parameter features, blood glucose trend features, and current physiological parameters to achieve accurate extraction and efficient fusion of multi-dimensional features.
[0040] See Figure 6 As shown, the blood glucose detection model includes: The first processing network is used to extract time-series features related to the current blood glucose value from historical physiological parameters and generate historical multi-parameter features. The second processing network is used to analyze the historical variation patterns of blood glucose parameters and generate blood glucose trend characteristics. A fusion network is used to fuse historical multi-parameter features, blood glucose trend features, and current physiological parameters to generate a fusion feature vector, and then transform the fusion feature vector into a blood glucose prediction value through a multi-layer fully connected network.
[0041] The blood glucose detection model extracts predicted blood glucose values based on a dual GRU network, specifically: Using physiological and blood glucose parameters from a historical time-series database as input, two GRU networks learn the patterns of blood glucose fluctuations at different time periods and extract historical trend features, covering key information such as the rate of blood glucose change, fluctuation cycle, and peak / trough occurrence patterns, accurately capturing the mechanism by which historical data affects current blood glucose. Specifically, the first processing network uses the output of the last hidden state, containing comprehensive information from multiple historical parameters; the second processing network also uses the output of the last hidden state, focusing on blood glucose change trends.
[0042] Furthermore, the first processing network, the second processing network, and the fusion network adopt GRU1, GRU2, and FCNN, respectively, with GRU1 and GRU2 representing two independent or parallel GRU modules. Specifically, in blood glucose monitoring, GRU1 is used to extract time-series features related to the current blood glucose value from historical physiological parameters to generate historical multi-parameter features; GRU2 is used to analyze the historical changes in blood glucose parameters and generate blood glucose trend characteristics. FCNN is a fully connected neural network, a three-dimensional feature fusion and blood glucose inference model. It is used to weightedly fuse historical multi-parameter features, blood glucose trend features, and current physiological parameters to generate a fused feature vector. A non-linear mapping of a three-layer fully connected network establishes the correlation between the fused feature vector and the blood glucose value, ultimately outputting an accurate blood glucose value. Specifically: The first fully connected layer uses the first ReLU activation function and Dropout regularization. The expression for the first ReLU activation function is: ; The second fully connected network uses the second ReLU activation function, which is expressed as follows: ; in , The output of the fully connected network is ReLU, where ReLU() is the activation function. , For the weight matrix of a fully connected network, , This is the bias vector corresponding to the fully connected network. To fuse feature vectors.
[0043] The third layer is a linear output layer, which obtains the predicted blood glucose value through a blood glucose output function. The predicted blood glucose value is a one-dimensional numerical value, and the specific formula is as follows: ; in This is a predicted blood glucose value. This is the output of the second-layer fully connected network. This is the bias scalar for a fully connected layer 3 network.
[0044] In some embodiments, the hidden layer dimensions of the three-layer fully connected network are fc_hidden_size, fc_hidden_size / / 2, and 1, respectively.
[0045] In some embodiments, historical multi-parameter features, blood glucose trend features, and current physiological parameters are weighted and fused to generate a fused feature vector, specifically including: First, historical physiological parameters are input into the GRU encoder to obtain historical multi-parameter features characterizing long-term dynamic patterns. Second, historical blood glucose parameters are statistically and trend-analyzed and mapped to blood glucose trend features through a fully connected layer. Meanwhile, the physiological parameters at the current moment are mapped to an instantaneous state vector h_curr through another fully connected layer. Then, a dynamic weighted fusion method based on attention mechanism is adopted to automatically calculate the importance weights of the three according to the current context and perform weighted summation to obtain a preliminary fusion feature vector. Finally, this vector is passed through a refinement network to generate a fully information-complete fusion feature vector for final prediction.
[0046] Furthermore, in some embodiments, a multi-parameter processing model is integrated, wherein the multi-parameter processing micro-model consists of an MCU, a Bluetooth module, and a power module. The MCU is based on the STM32's ARM Cortex-M4 core and provides rich external interfaces to coordinate signal acquisition, data processing, and real-time transmission; the Bluetooth module uses a BLE2U-A chip and peripheral circuitry to realize wireless data interaction between the device and the host via the UART protocol; the power module converts the 3.7V lithium-ion battery voltage to a stable 3.3V power supply through a low-dropout linear regulator (LDO, model ME6209A33).
[0047] To achieve the above objectives, another aspect of this application proposes a non-invasive / minimally invasive blood glucose prediction method, wherein the method includes: Collect blood glucose and physiological parameters and input them into the blood glucose detection model; The model outputs predicted blood glucose values based on the blood glucose detection model.
[0048] See Figure 7 As shown, in some embodiments, the output of the blood glucose prediction value by the blood glucose detection model specifically includes, but is not limited to, steps S101-S103, as follows: S101: Extract time-series features related to the current blood glucose level from historical physiological parameters to generate historical multi-parameter features; S102: Analyze the historical variation patterns of blood glucose parameters and generate blood glucose trend characteristics; S103: The historical multi-parameter features, blood glucose trend features, and current physiological parameters are fused to generate a fused feature vector, and the predicted blood glucose value is output based on the fused feature vector.
[0049] In some embodiments, step S102 involves analyzing the historical variation patterns of blood glucose parameters and generating blood glucose trend characteristics, specifically as follows: Input the blood glucose parameters from the previous period into the sliding window, and calculate the overall trend characteristics of all blood glucose parameters within the sliding window; Input the blood glucose parameters of the previous period into the sliding window, divide the sliding window into several sub-windows, calculate the average blood glucose concentration of each sub-window, and calculate the average fluctuation characteristics based on the average blood glucose concentration of all sub-windows. The overall trend features and the average trend features are weighted and averaged to generate blood glucose trend features.
[0050] Specifically, since blood glucose fluctuations have the characteristics of both short-term minute changes and long-term major trends, in order to fully capture their dynamic change patterns, it is necessary to combine sliding windows of different time scales and two calculation methods, and finally achieve a stable representation of blood glucose fluctuations through comprehensive indicators.
[0051] See Figure 5 As shown, this computational framework uses a sliding time window to assess background dynamics. Through scanning time windows ranging from 15 to 90 minutes, it integrates average fluctuations at different time scales, providing a reliable characterization of both small and significant fluctuations for each blood glucose measurement. Specifically, the calculation principle of the baseline blood glucose trend #1 considers the overall fluctuations among all data points within the sliding time window. This is achieved by calculating the average BG trend value of each data point within the sliding window (subtracting the BG value from the previous time point from the BG value at a later time point, and then dividing by the time difference between the two data points), and then performing a weighted average of the trend values of all data points within the sliding window, thus obtaining the overall trend #1 of the blood glucose data. The calculation principle of baseline blood glucose trend #2 considers the average trend of blood glucose data points in the first and second halves of the sliding window. The average BG value for the first and second halves of the sliding window is calculated separately. Then, the result of subtracting the average BG value for the second half from the first half's average BG value is divided by the average time difference between the two halves to obtain the overall trend #2. Finally, a weighted average of baseline blood glucose trends #1 and #2 from different time windows is performed to integrate the major and minor trends.
[0052] Specifically, in this embodiment, different sliding window sizes of 15 minutes, 30 minutes, 45 minutes, 60 minutes, 75 minutes, and 90 minutes are used. The trend calculation for each blood glucose parameter point uses the corresponding time point as the window center. The trend value is calculated using two independent methods, and then averaged to obtain the final blood glucose trend value for that time point. In practical applications, the window size can be flexibly adjusted according to the time interval of blood glucose monitoring. The two trend calculation methods are the overall fluctuation trend method within the window and the average trend method before and after the window, specifically: The overall fluctuation trend method within a window: The core is to capture the overall fluctuation characteristics of all data points within a sliding window. Taking a 15-minute window as an example, when calculating the trend value at a certain time t2, first construct a 15-minute window centered on t2, calculate the trend value of adjacent data points within the window (such as t1 and t2, t3 and t4) (the BG value of the later time point minus the BG value of the previous time point, then divided by the 5-minute time difference), and finally perform a weighted average of the trend values of all adjacent data points to obtain the overall trend value at that time.
[0053] The average trend method for the first and second halves of a sliding window focuses on the difference in average trends between the two halves. Taking a 15-minute window as an example, to calculate the trend value at time t2, first construct a window centered at t2, then calculate the average BG value for the first and second halves of the window. Finally, subtract the average BG value for the second half from the average BG value for the first half, and divide the result by the average time difference between the two halves to obtain the trend value at that moment.
[0054] Finally, the blood glucose trend values from the two methods were weighted and averaged to integrate the average blood glucose fluctuation characteristics across different time ranges. This approach can reflect both short-term minor fluctuations and long-term major trends, providing a comprehensive and stable basis for representing fluctuations in each blood glucose measurement.
[0055] The technical solution of this application will be explained below with specific application examples: The model was trained and tested using a 5-day minimally invasive / non-invasive CGM dataset (approximately 5 hours of recording per day) from 18 participants (9 diabetic patients and 9 healthy individuals). By inputting the blood glucose parameters and physiological parameters from the test set into the trained blood glucose detection model, non-invasive derivation of blood glucose values can be achieved.
[0056] See Figure 8 As shown in the embodiments of this application, the Clarke error grid analysis of BG values during the initial training of the blood glucose detection model was performed on the diabetic group (DP#1 to DP#9) and the healthy group (HP#1 to HP#9). Most of the blood glucose values for each subject in the training dataset were distributed in region A (corresponding to a detection error <20%), meeting the clinical requirement of an error <15%.
[0057] See Figure 9 As shown, this application presents a Clarke error grid analysis of blood glucose trends during the initial model training for the diabetes group (DP#1 to DP#9) and the healthy group (HP#1 to HP#9). The vast majority of blood glucose trend data for each subject in the training dataset is distributed in region A.
[0058] See Figure 10 The results of Clarke error grid analysis of the accuracy of blood glucose levels and trends in diabetic patients and healthy subjects during the model testing period provided in this application are shown. Clarke error grid analysis shows that the blood glucose distribution range in the diabetic patient group is in good agreement with that in the healthy control group. More than 90% of the data in the test set is distributed in zone A (error <15%), with only a small portion distributed in zone B, fully validating the clinical reliability of the monitoring results. Regarding blood glucose trends, 100% of the trend data for most participants is located in zone A. Overall, there is a certain difference in accuracy between the blood glucose monitoring model and the healthy group, but this difference is within a reasonable range. The average MARD value for all participants is approximately 15.6%, which meets the reasonable requirements for CGM monitoring.
[0059] To achieve the above objectives, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.
[0060] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.
[0061] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.
[0062] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.
[0063] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.
[0064] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.
[0065] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.
[0066] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.
[0067] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0068] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.
[0069] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.
[0070] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.
[0071] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.
[0072] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0073] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.
Claims
1. A non-invasive / minimally invasive blood glucose prediction system, characterized in that, The system includes: The data acquisition module is used to collect blood glucose and physiological parameters through non-invasive and minimally invasive methods and input them into the blood glucose detection model; The blood glucose prediction module is used to output a predicted blood glucose value through the blood glucose detection model.
2. The non-invasive / minimally invasive blood glucose prediction system according to claim 1, characterized in that, The data acquisition module includes: Microneedle electrode unit is used to collect blood glucose parameters in tissue fluid; The non-invasive blood glucose sensing unit is used to collect physiological parameters through a sensor array, including historical physiological parameters and current physiological parameters.
3. The non-invasive / minimally invasive blood glucose prediction system according to claim 2, characterized in that, The blood glucose prediction module includes A first processing network is used to extract time-series features related to the current blood glucose value from the historical physiological parameters and generate historical multi-parameter features. The second processing network is used to analyze the historical variation patterns of the blood glucose parameters and generate blood glucose trend characteristics. A fusion network is used to fuse the historical multi-parameter features, blood glucose trend features, and current physiological parameters to generate a fusion feature vector, and then transform the fusion feature vector into a blood glucose prediction value through a multi-layer fully connected network.
4. The non-invasive / minimally invasive blood glucose prediction system according to claim 3, characterized in that, The converged network comprises several layers of fully connected networks, including: The first ReLU activation function is expressed as follows: ； The second ReLU activation function is expressed as follows: ； The blood glucose output function, the expression of which is: ； in , For the output of a fully connected network, Here, ReLU is the predicted blood glucose value, and ReLU() is the activation function. , as well as For the weight matrix of a fully connected network, , as well as This is the bias vector corresponding to the fully connected network. To fuse feature vectors.
5. The non-invasive / minimally invasive blood glucose prediction system according to claim 2, characterized in that, The data acquisition module also includes: The MCU module is used to convert the signals collected by the microneedle electrode unit and the non-invasive blood glucose sensor unit into blood glucose parameters and physiological parameters, respectively. Data storage module, used to store the blood glucose parameters and physiological parameters; A wireless transmission module is used to transmit the blood glucose parameters and physiological parameters to the blood glucose detection model; Power module, used to provide power.
6. The non-invasive / minimally invasive blood glucose prediction system according to claim 2, characterized in that, The sensor group includes a temperature sensor, an infrared radiation sensor, and a photoelectric sensor.
7. The non-invasive / minimally invasive blood glucose prediction system according to claim 1, characterized in that, The step of outputting a predicted blood glucose value through the blood glucose detection model specifically includes: Extract time-series features related to current blood glucose levels from historical physiological parameters to generate historical multi-parameter features; Analyze the historical patterns of blood glucose parameters to generate blood glucose trend characteristics; The historical multi-parameter features, blood glucose trend features, and current physiological parameters are fused to generate a fused feature vector, which is then transformed into a blood glucose prediction value through a multi-layer fully connected network.
8. The non-invasive / minimally invasive blood glucose prediction system according to claim 7, characterized in that, The analysis of historical changes in blood glucose parameters and the generation of blood glucose trend characteristics specifically include: Input the blood glucose parameters from the previous period into the sliding window, and calculate the overall trend characteristics of all blood glucose parameters within the sliding window; Input the blood glucose parameters of the previous period into the sliding window, divide the sliding window into several sub-windows, calculate the average blood glucose concentration of each sub-window, and calculate the average fluctuation characteristics based on the average blood glucose concentration of all sub-windows. The overall trend features and the average trend features are weighted and averaged to generate blood glucose trend features.
9. A non-invasive / minimally invasive method for predicting blood glucose levels, characterized in that, The method includes: Blood glucose and physiological parameters are collected through non-invasive and minimally invasive methods and then input into the blood glucose detection model; The blood glucose detection model outputs a predicted blood glucose value.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of claim 9.