Diabetic neuropathy dynamic early warning method based on multi-modal biological signals

Abstract

The application provides a kind of diabetic neuropathy dynamic early warning method based on multi-modal biological signal, through the integration of multi-modal sensor of intelligent wearable device, combine deep learning algorithm to construct individualized prediction model, realize the risk prediction of diabetic neuropathy (DN) progression and real-time warning of unconscious hypoglycemia through artificial intelligence algorithm.Can according to the time-frequency domain characteristics of multi-modal signal and clinical data, predict the time window of DN progression to serious stage;Through ECG and HR characteristics combined with blood glucose trend prediction model, identify the risk of unconscious hypoglycemia, and trigger graded intervention measures;Based on the physiological response mode of individual patient, dynamically optimize the threshold of algorithm, improve the accuracy of prediction and early warning.The application integrates HR, BP and ECG and other multi-dimensional signals, improves the prediction accuracy;Support continuous monitoring and instant feedback;Through adaptive algorithm to adapt to individual differences of patients, overcome the limitations of traditional general model.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomedical engineering and intelligent wearable devices, in particular to a dynamic early warning method for diabetic neuropathy based on multi-modal biological signals, a dynamic early warning device for diabetic neuropathy based on multi-modal biological signals, an electronic device and a computer readable storage medium. BACKGROUND

[0002] Diabetic neuropathy (DN) is one of the most common chronic complications of diabetes, and its early diagnosis and disease monitoring mainly rely on invasive examination methods such as nerve conduction test and quantitative sensory test. In the current clinical practice, the early warning system for asymptomatic hypoglycemia and neuropathy progression risk is mainly based on the following technical framework:

[0003] Single-dimensional biological signal monitoring: using continuous glucose monitoring system (CGMS) or discrete ECG recorder, the risk is judged through single glucose fluctuation curve or fixed threshold heart rate variability (HRV) index. Such systems rely on 30-second ECG segment analysis and blood glucose standard deviation calculation, and lack time synchronization of multiple physiological dimensions.

[0004] Static risk assessment model: using Logistic regression or traditional machine learning algorithm (such as random forest), a grading prediction model is constructed based on quarterly clinical examination data (such as HbA1c, ankle-brachial index). The CAN risk score recommended by the American Diabetes Association (ADA) guidelines belongs to this category, and its parameter update cycle is as long as 3-6 months.

[0005] Manual threshold alarm mechanism: the existing intelligent wearable devices trigger hypoglycemia warning by setting fixed thresholds (such as heart rate <50bpm for 5 minutes), and the parameter setting is based on group statistical characteristics rather than individual physiological baseline.

[0006] However, the above-mentioned traditional prior art still has the following application defects:

[0007] Multi-modal data mismatch risk: discrete monitoring devices cause time asynchrony (average deviation >120 seconds) in ECG, blood pressure, blood glucose and other signals, especially in patients with autonomic nervous dysfunction, the cardiovascular compensation reaction lags behind the blood glucose change (lag coefficient β=0.32), which will significantly reduce the prediction sensitivity.

[0008] Dynamic pathological characterization loss: the time domain features (such as SDNN) relied on by traditional models cannot capture the nonlinear dynamic characteristics specific to diabetic cardiac autonomic neuropathy. Studies have shown that when the sympathetic / parasympathetic nerve regulation ratio (LF / HF) >3.5, the prediction accuracy of low blood sugar based on FFT-based spectral analysis method decreases by 37%.

[0009] Individual adaptability defects: The fixed threshold mechanism ignores the autonomic nerve remodeling characteristics of DN patients (such as a 40% down-regulation of β2 adrenergic receptors), resulting in a decrease in specificity. Clinical data shows that the standard alarm threshold has a false positive rate of up to 68% in patients with a disease duration of >10 years.

[0010] Human intervention delay: Existing systems require medical staff to manually calibrate model parameters (an average period of 21 days), which cannot respond in real time to non-linear physiological changes caused by acute metabolic disorders. In the case of nocturnal asymptomatic hypoglycemia events, manual review delays result in an effective intervention rate of less than 42%.

[0011] Cross-modal correlation breakage: Current algorithms separately process blood glucose trends and cardiovascular signals, failing to establish a dynamic coupling model between the two. In particular, when autonomic neuropathy causes a decrease in baroreflex sensitivity (BRS) to <3 ms / mmHg, the area under the ROC curve (AUC) of a single modality prediction model decreases significantly by 0.15-0.22.

[0012] These defects result in a prediction error range of ±18 months for DN progression risk in the prior art, a warning sensitivity of only 59%-67% for unconscious hypoglycemia, and a nocturnal false negative rate of up to 73% of the total system error. In particular, the human-dependent calibration mechanism makes it impossible for the prediction model to adapt to the neural plasticity changes of DN patients in a timely manner, which has become a key bottleneck restricting accurate early warning. SUMMARY

[0013] To solve the technical problems existing in the prior art, the present application provides the following technical solutions:

[0014] On the one hand, a dynamic early warning method for diabetic neuropathy based on multi-modal biological signals is provided, which is realized by an electronic device, and the method comprises:

[0015] S1, real-time acquisition of biological signal data of DN patients: heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG), and input into a pre-set preprocessing engine;

[0016] S2, pre-processing of the biological signal data by the preprocessing engine, and real-time extraction and acquisition of feature data of DN patients: heart rate variability parameters (HRV), blood pressure variability parameters (BPV) and electrocardiogram morphology parameters (ECG-v);

[0017] S3, inputting the feature data into a pre-deployed DN dynamic early warning model, identifying the feature data by the model and outputting corresponding DN risk prediction results, including DN progression probability and hypoglycemia risk index;

[0018] S4, determining whether to trigger an alarm according to the risk prediction results, and outputting a corresponding determination result;

[0019] S5, record the biological signal data, feature data, risk prediction result and judgment result of the DN patient, and store and upload to the cloud.

[0020] Preferably, in step S1, the biological signal data of the DN patient is collected in real time in the following manner:

[0021] The biological signal data of the DN patient is collected by a smart wearable device worn by the DN patient, including heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG);

[0022] And,

[0023] The motion artifact correction algorithm is used to correct the motion artifact of the biological signal data, and the corrected data is uploaded to the preprocessing engine.

[0024] Preferably, in step S3, the DN early warning model is generated by the following steps:

[0025] Obtain a plurality of biological signal data sets of DN patients, including multi-modal biological signals at different time sequences: heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG), and label data: DN severity classification and continuous glucose monitoring data (CGM);

[0026] The preprocessing engine is used to preprocess the biological signal data of each DN patient and extract corresponding feature data in real time: heart rate variability parameter (HRV), blood pressure variability parameter (BPV) and electrocardiogram morphology parameter (ECG-v);

[0027] According to the DN severity classification, the feature data is labeled with the corresponding DN progression probability label;

[0028] According to the continuous glucose monitoring data (CGM), the feature data is labeled with the corresponding hypoglycemia risk index label;

[0029] The feature data of each DN patient is counted to obtain a data set for training the model, and the data set is divided into a training set, a validation set and a test set according to a proportion;

[0030] The training set is input into a preset LSTM-CNN model for feature learning to generate an initial DN early warning model;

[0031] The validation set is used to verify the application performance of the DN early warning model:

[0032] If the verification is passed, the test set is used to test the prediction performance of the DN early warning model: if it is passed, the DN early warning model is deployed and applied; otherwise, it is retrained;

[0033] Conversely, retrain.

[0034] Preferably, in step S4, according to the risk prediction result, it is judged whether to trigger an alarm, and the corresponding judgment result is output, including:

[0035] The pre-warning threshold of the DN patient on the DN progression probability and the hypoglycemia risk index is set according to the individual data of the DN patient, including:

[0036] Based on the difference between the HRV / BPV / ECG-v distribution of healthy people and DN patients, the basic threshold is determined by kernel density estimation;

[0037] Through the individual historical data of the DN patient: the trend of HRV / BPV or the abnormal frequency of ECG-v, the threshold is dynamically adjusted by Bayesian optimization:

[0038] ,

[0039] Wherein:

[0040] The basic threshold is;

[0041] The current threshold is;

[0042] The optimized threshold is;

[0043] η is the set learning rate;

[0044] The set log-likelihood gradient in the Bayesian function is;

[0045] The risk prediction result of the DN patient is obtained, including the DN progression probability and the hypoglycemia risk index of the DN patient;

[0046] It is judged whether the DN progression probability of the DN patient exceeds the pre-warning threshold of the DN patient on the DN progression probability:

[0047] If yes, the corresponding probability DN clinical examination strategy is generated; otherwise, it is abandoned;

[0048] It is judged whether the hypoglycemia risk index of the DN patient exceeds the pre-warning threshold of the DN patient on the hypoglycemia risk index:

[0049] If yes, an alarm is triggered, and the corresponding alarm signal of the DN patient is generated; otherwise, it is abandoned.

[0050] On the other hand, a dynamic warning device for diabetic neuropathy based on multi-modal biological signals is provided, which is applied to a dynamic warning method for diabetic neuropathy based on multi-modal biological signals, and the device comprises:

[0051] The intelligent wearable device is used for collecting biological signal data of DN patients in real time, including heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG), and inputting the biological signal data into a preset preprocessing engine.

[0052] The preprocessing engine is used for preprocessing the biological signal data, extracting and acquiring feature data of DN patients in real time, including heart rate variability parameters (HRV), blood pressure variability parameters (BPV) and electrocardiogram morphology parameters (ECG-v), and inputting the feature data into a prewarning system.

[0053] The prewarning system is used for inputting the feature data into a DN dynamic prewarning model deployed in advance, identifying the feature data by the model and outputting corresponding DN risk prediction results, including DN progression probability and hypoglycemia risk index.

[0054] The dynamic threshold module is used for judging whether the risk prediction results trigger an alarm according to preset prewarning thresholds of DN patients on DN progression probability and hypoglycemia risk index respectively, and outputting corresponding judgment results.

[0055] The data service module is used for recording biological signal data, feature data, risk prediction results and judgment results of DN patients, and storing and uploading the data to the cloud.

[0056] The intelligent wearable device (such as a smart watch) is integrated with sensors for collecting various modalities, can be implemented and understood in combination with existing intelligent wearable devices, and can transmit data to a background server wirelessly.

[0057] In another aspect, an electronic device is provided, which includes a processor, and a memory having computer readable instructions stored thereon, the computer readable instructions, when executed by the processor, implement any one of the above methods for dynamic prewarning of diabetic neuropathy based on multi-modal biological signals.

[0058] In another aspect, a computer readable storage medium is provided, which stores at least one instruction, the at least one instruction is loaded and executed by a processor to implement any one of the above methods for dynamic prewarning of diabetic neuropathy based on multi-modal biological signals.

[0059] The technical scheme provided by the embodiments of the present application has at least the following beneficial effects:

[0060] The application provides a diabetic neuropathy dynamic early warning system based on multi-modal biological signals. The system integrates multi-modal sensors through an intelligent wearable device (such as a smart watch), collects signals such as heart rate (HR), blood pressure (BP), and single-lead electrocardiogram (ECG), and constructs a personalized prediction model combined with a deep learning algorithm. The system realizes diabetic neuropathy (DN) progression risk prediction and real-time warning of unconscious hypoglycemia through artificial intelligence algorithms. The system can achieve the following goals:

[0061] Dynamic risk stratification: According to the time-frequency domain features of multi-modal signals and clinical data, the time window for DN progression to a severe stage is predicted.

[0062] Key indicator warning: Through the combination of ECG and HR features and blood glucose trend prediction models, the risk of unconscious hypoglycemia is identified, and graded intervention measures are triggered.

[0063] Adaptive learning mechanism: Based on the individual physiological response patterns of patients, the algorithm threshold is dynamically optimized to improve the accuracy of prediction and warning.

[0064] The core advantages of the application are:

[0065] Multi-modal fusion: Integrating multi-dimensional signals such as HR, BP, and ECG improves prediction accuracy.

[0066] Real-time: Supports continuous monitoring and immediate feedback.

[0067] Personalization: Through adaptive algorithms, individual differences are adapted to overcome the limitations of traditional general models. BRIEF DESCRIPTION OF DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0069] Figure 1 is a dynamic early warning method flowchart for diabetic neuropathy based on multi-modal biological signals provided by the embodiments of the application;

[0070] Figure 2 is a training mechanism schematic diagram of a DN early warning model provided by the embodiments of the application;

[0071] Figure 3 is a dynamic early warning device block diagram for diabetic neuropathy based on multi-modal biological signals provided by the embodiments of the application;

[0072] Figure 4 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0073] The technical solutions in the present application will be described below with reference to the drawings.

[0074] In the embodiments of the present application, the words such as "example", "for example" are used to represent as an example, illustration or description. Any embodiment or design scheme described as "example" in the present application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. Rather, the word "example" is intended to present the concept in a specific manner. In addition, in the embodiments of the present application, the meaning expressed by "and / or" can be both, or can be one of the two.

[0075] In the embodiments of the present application, "image" and "picture" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent. "Of", "corresponding" and "corresponding" can be used interchangeably at times. It should be pointed out that when the distinction is not emphasized, the meanings expressed are consistent.

[0076] In the embodiments of the present application, sometimes the subscript such as W1 may be mistakenly used in the form of non-subscript such as W1. When the distinction is not emphasized, the meanings expressed are consistent.

[0077] In order to make the technical problems, technical solutions and advantages to be solved by the present application more clear, the following will be described in detail with reference to the drawings and specific embodiments.

[0078] The embodiments of the present application provide a dynamic early warning method for diabetic neuropathy based on multi-modal biological signals. The method can be realized by an electronic device, which can be a terminal or a server. As shown in the flow chart of the dynamic early warning method for diabetic neuropathy based on multi-modal biological signals, the processing flow of the method can include the following steps: Figure 1

[0079] S1, real-time acquisition of biological signal data of DN patients: heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG), and input into a preset preprocessing engine;

[0080] S2, preprocessing of the biological signal data by the preprocessing engine, and real-time extraction and acquisition of feature data of DN patients: heart rate variability parameter (HRV), blood pressure variability parameter (BPV) and electrocardiogram morphology parameter (ECG-v);

[0081] ​S3. Inputting the characteristic data into a pre-deployed DN dynamic early warning model, which recognizes the characteristic data and outputs corresponding DN risk prediction results, including DN progression probability and hypoglycemia risk index;

[0082] S4. Determine whether to trigger an alarm based on the risk prediction result, and output a corresponding determination result;

[0083] S5. Record the biological signal data, characteristic data, risk prediction results and judgment results of DN patients, store them and upload them to the cloud.

[0084] Combined with attachment Figure 3 The present invention primarily integrates multimodal sensors into smart wearable devices (such as smartwatches) to collect signals such as heart rate (HR), blood pressure (BP), and single-lead electrocardiogram (ECG). This is combined with deep learning algorithms to build a personalized prediction model to predict the probability of DN progression and hypoglycemia risk index in DN patients. Furthermore, the ECG and HR features are combined with a blood glucose trend prediction model to identify the risk of unaware hypoglycemia and trigger graded intervention measures.

[0085] At the same time, the system has an adaptive learning mechanism: based on the patient's individual physiological response pattern, it dynamically optimizes the algorithm threshold to improve the accuracy of prediction and warning.

[0086] Hospitals can distribute smart wearable devices to DN patients and create a backend file, binding the device ID and patient medical record number, etc., so that after the device is activated, the data reported by the device can be written to the patient's electronic medical record. The backend deploys a preprocessing engine, an early warning system, a dynamic threshold module, and a system output module (specifically, in conjunction with the device described in this application), which can realize corresponding data analysis and intelligent analytical calculation and judgment with the computing power of the CPU processing system.

[0087] This will be described in detail below.

[0088] Preferably, in step S1, the method of collecting biological signal data of DN patients in real time is as follows:

[0089] The biosignal data including heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG) are collected through smart wearable devices worn by DN patients;

[0090] as well as,

[0091] The biosignal data is subjected to motion artifact correction using a motion artifact correction algorithm, and the corrected data is uploaded to the preprocessing engine.

[0092] The intelligent wearable device is integrated with various sensors, power supply, MCU chip, memory, processor and wireless module (for reference, the existing smart bracelet, etc.), and can collect multi-modal biological signals such as HR, BP and ECG.

[0093] The intelligent wearable device is integrated with multi-modal sensors: for example, a photoplethysmography (PPG) sensor is used to measure HR, an oscillograph is used to measure BP, and the sampling rate is ≥200Hz; a single-lead ECG electrode is integrated to collect electrocardiogram signals (an accelerometer can also be integrated for motion monitoring correction); a processor is used for signal preprocessing such as A / D conversion, and a built-in motion artifact correction algorithm is used to ensure the stability of the signal in a dynamic environment.

[0094] The motion artifact correction algorithm can correct the motion artifact of the sampled signal, improve the accuracy of the signal feature, and facilitate better feature recognition by the model. For example, in order to solve the problems of PPG waveform distortion (such as peak inversion) and ECG baseline drift caused by motion artifacts in traditional methods, the following motion artifact correction algorithm is adopted:

[0095] 1. Multi-modal time synchronization correction:

[0096] ,

[0097] is the multi-modal signal after time synchronization (including PPG, ECG and BP);

[0098] S i is the i-th modal signal;

[0099] is the inherent delay of the i-th sensor (for example, PPG: 50ms, ECG: 20ms, BP: 100ms);

[0100] is the Dirac impulse function, used for timestamp alignment;

[0101] is a dynamic compensation factor (calculated based on accelerometer data).

[0102] 2. Motion noise separation correction:

[0103] ,

[0104] is the motion artifact component;

[0105] is the three-axis accelerometer signal (sampling rate ≥200Hz);

[0106] The wavelet packet decomposition weight matrix (Daubechies-4 base function) is as follows: A 4-layer decomposition is adopted to extract the 0.5-10 Hz frequency band (corresponding to human motion frequency), and the formula is:

[0107] ,

[0108] The j-th layer wavelet base function is as follows:

[0109] The energy normalization coefficient (C1=0.6, C2=0.3, C3=0.08, C4=0.02) is as follows:

[0110] The convolution operation is used to extract the frequency domain coupling feature.

[0111] The independent component analysis is used to separate the signal and noise source.

[0112] 3. Adaptive filter compensation:

[0113] ,

[0114] The corrected output signal is as follows:

[0115] The dynamic attenuation coefficient (range 0.2-1.0, controlled by the signal SNR) is as follows:

[0116] k is an empirical constant, and the preferred value is k=0.5. SNR is calculated by the ratio of PPG signal AC / DC components, and the filter strength is dynamically adjusted;

[0117] The nonlinear activation function suppresses low-frequency motion interference.

[0118] Through multi-modal signal fusion and dynamic parameter adjustment, the problems of PPG waveform distortion (such as peak inversion) and ECG baseline drift caused by motion artifacts in traditional methods are solved, and the measurement error of heart rate and other signals is reduced.

[0119] The signal is uploaded to the background, and the pre-processing engine of the background performs pre-processing and extracts signal features.

[0120] The pre-processing engine function: clean and extract features from the original signal. Real-time band-pass filtering (0.5-40 Hz) to remove noise and baseline drift. Extract features, such as: R-wave detection algorithm to extract QRS complex, calculate heart rate variability (HRV); wavelet packet transform to calculate multi-scale entropy of HRV, extract time-frequency domain features. Specifically:

[0121] 1. Preprocessing Engine Core Workflow‌

[0122] Signal Denoising and Baseline Correction‌

[0123] Adaptive Filtering: Employing Kalman filter combined with wavelet domain decomposition to eliminate motion artifacts and high-frequency noise. For ECG signals, a 0.05-40Hz band-pass filter is selected to suppress electromyographic interference and power frequency noise.

[0124] Baseline Drift Correction: Eliminate low-frequency baseline drift of ECG signals through Cubic Spline Interpolation, ensuring the stability of QT interval measurement.

[0125] Multi-modal Time Synchronization: Utilize accelerometer data to compensate for the time deviation (±50ms) between PPG and ECG, ensuring the time alignment of HRV and BPV parameters.

[0126] Signal Segmentation and Sliding Window Mechanism‌

[0127] Dynamic Window Partitioning: Dynamically adjust the window length based on R-wave intervals (default 2 seconds), adapting to the signal period characteristics at different heart rates.

[0128] Overlap Sampling Strategy: Window overlap rate of 50% to improve the continuity of short-term HRV parameters (such as SDNN) calculation.

[0129] 2. Heart Rate Variability (HRV) Parameter Extraction‌

[0130] R-wave Detection and QRS Complex Localization‌

[0131] Difference Threshold Algorithm: Combining first-order difference and adaptive threshold to detect R-wave peaks, with better anti-interference ability than traditional Pan-Tompkins algorithm.

[0132] Morphological Correction: Recognize QRS onset and termination points through second derivative method, eliminating false detection caused by high T-wave.

[0133] Multi-scale Entropy and Time-frequency Domain Feature Calculation‌

[0134] Wavelet Packet Decomposition: Using Daubechies-4 wavelet basis for 5-layer decomposition, extracting the energy proportion of 0.04-0.4Hz (corresponding to sympathetic / parasympathetic nerve activity frequency band).

[0135] Nonlinear Dynamics Parameters: Calculate Sample Entropy and Multifractal Spectrum to quantify the chaotic characteristics of autonomic nervous regulation.

[0136] Typical HRV Parameters‌

[0137] Time-domain parameters: SDNN (standard deviation), RMSSD (root mean square of successive differences), PNN50 (proportion of adjacent RR intervals with a difference > 50 ms).

[0138] Frequency-domain parameters: LF / HF ratio (low / high frequency energy ratio, reflecting sympathetic-parasympathetic balance), total power (TP).

[0139] 3. Blood pressure variability (BPV) parameter extraction

[0140] Blood pressure time series preprocessing

[0141] Trend removal: separate the slow trend of blood pressure from the rapid fluctuation component by moving average filtering (window length 30 seconds).

[0142] Outlier correction: eliminate blood pressure spikes caused by motion artifacts based on median filtering (window length 5 samples).

[0143] BPV feature quantification method

[0144] Time-frequency analysis: use continuous wavelet transform (CWT) to calculate the energy density distribution of blood pressure fluctuations in the 0.01-0.15 Hz frequency band.

[0145] Statistical features: blood pressure standard deviation (BP-SD), coefficient of variation (CV=SD / mean x 100%), dynamic trend slope (DTS).

[0146] 4. Electrocardiogram morphology parameter (ECG-v) extraction

[0147] Key waveform recognition algorithm

[0148] QT interval measurement: locate the T wave end based on the dynamic threshold method, and eliminate the influence of heart rate by combining the heart rate correction formula.

[0149] ST segment analysis: sample the ST segment slope at 80 ms after J point, and quantify ischemic ST segment depression / elevation by polynomial fitting.

[0150] T wave symmetry index: calculate the time ratio of the ascending branch and the descending branch of the T wave, with a normal range of 0.88-1.12.

[0151] Morphological dynamic tracking

[0152] Dynamic time warping (DTW): compare the morphological differences between the current heartbeat and the template waveform, and quantify the myocardial repolarization dispersion.

[0153] Covariance matrix analysis: construct the covariance matrix features of QRS-T wave groups to detect small repolarization abnormalities.

[0154] 5. Real-time computing architecture optimization

[0155] Heterogeneous parallel processing: ECG feature extraction and HRV calculation are assigned to FPGA hardware acceleration module, and BPV analysis is performed by ARM Cortex-M4 core, with delay reduced to 28 ms.

[0156] Memory optimization strategy: Ring Buffer is used to store sliding window data, reducing memory occupancy by 40%.

[0157] Model lightweight: GRU network parameter quantity is compressed to 50 KB, supporting real-time operation on wearable devices.

[0158] By integrating time-frequency analysis (such as wavelet packet transform), nonlinear dynamics (such as multiscale entropy), and morphological quantification (such as QT interval dynamic tracking), the accuracy of autonomic nervous function evaluation for diabetic neuropathy (DN) patients is significantly improved.

[0159] As shown in Figure 2 , preferably, in step S3, the method for generating the DN early warning model comprises the following steps:

[0160] Obtain a plurality of DN patient biosignal data sets, including multi-modal biosignals at different time sequences: heart rate (HR), blood pressure (BP), and single-lead electrocardiogram (ECG), as well as label data: DN severity classification and continuous glucose monitoring data (CGM);

[0161] Preprocess the biosignal data of each DN patient through the preprocessing engine, and extract the corresponding feature data in real time: heart rate variability parameters (HRV), blood pressure variability parameters (BPV), and electrocardiogram morphological parameters (ECG-v);

[0162] According to the DN severity classification, the feature data is labeled with the corresponding DN progression probability;

[0163] According to the continuous glucose monitoring data (CGM), the feature data is labeled with the corresponding hypoglycemia risk index;

[0164] Statistical analysis of the feature data of each DN patient to obtain a data set for training model, and dividing the data set into training set, validation set and test set according to the proportion;

[0165] Input the training set into the preset LSTM-CNN model for feature learning, and train to generate an initial DN early warning model;

[0166] Use the validation set to verify the application performance of the DN early warning model:

[0167] If the verification passes, test the prediction performance of the DN early warning model with the test set: if it passes, deploy and apply the DN early warning model; otherwise, retrain;

[0168] Otherwise, retrain.

[0169] Early warning system (early warning result calculation and analysis based on deployed algorithm chip) function: predict DN progression and hypoglycemia risk based on multi-modal signals. Hybrid model combining long short-term memory network (LSTM) and convolutional neural network (CNN):

[0170] Input parameters: time series data of standardized HRV, blood pressure variability (BPV), ECG morphological parameters (such as QT interval, ST segment changes);

[0171] Output results: DN progression probability (0-1) and hypoglycemia risk index (classification: low, medium, high).

[0172] Specific model training and application as follows:

[0173] 1. Data preparation

[0174] Input data: obtain time series data after cleaning and feature extraction from preprocessing engine, including:

[0175] Heart rate variability (HRV) parameters: such as multi-scale entropy, frequency domain features (LF / HF ratio, etc.).

[0176] Blood pressure variability (BPV) parameters: such as time series of blood pressure fluctuations.

[0177] Electrocardiogram morphological parameters (ECG-v): such as QT interval, ST segment changes, T wave symmetry, etc.

[0178] Data standardization: standardize different features to ensure consistent numerical ranges and avoid model training biases due to dimensional differences. Common methods include:

[0179] Z-score normalization: convert data to a distribution with mean 0 and standard deviation 1.

[0180] Min-Max scaling: scale data to the [0, 1] interval.

[0181] 2. Model architecture design

[0182] Model selection: hybrid model combining long short-term memory network (LSTM) and convolutional neural network (CNN):

[0183] LSTM: good at processing time series data, capturing long-term dependencies of signals such as HRV and BPV over time.

[0184] CNN: Extract spatial features of ECG signals, such as local morphological patterns of waveforms.

[0185] Model Architecture:

[0186] Input Layer: Receives standardized multi-modal time-series data with dimensions (sample number, time step, feature number).

[0187] CNN Layer: Applies one-dimensional convolutional layers to ECG signals, using multiple convolutional kernels to extract local features (e.g., QRS complex, T-wave morphology).

[0188] LSTM Layer: Inputs the features extracted by CNN and HRV, BPV time-series data into LSTM layers to learn the trend of signal changes over time.

[0189] Fusion Layer: Fuses the outputs of CNN and LSTM through a fully connected layer to generate comprehensive features (F) representation:

[0190] ,

[0191] Training Weight Matrix: Used to linearly weight the CNN output features C and LSTM output features L. WC measures the importance of spatial features extracted by CNN (e.g., ECG waveform abnormalities) to the fusion features, WL measures the contribution of time-series features extracted by LSTM (e.g., HRV circadian rhythm) to the fusion features, b (bias term) provides a baseline offset for the fusion features, and σ (ReLU activation function).

[0192] Output Layer:

[0193] DN Progress Prediction: Outputs a probability value between 0 and 1, representing the probability of DN progressing to a severe stage within a certain period of time.

[0194] Hypoglycemia Risk Prediction: Outputs a graded risk index (e.g., 0: low risk, 1: medium risk, 2: high risk).

[0195] 3. Model Training

[0196] Dataset: Use a dataset containing a large number of diabetic patient biological signal records, requiring:

[0197] Time-series data of multi-modal biological signals.

[0198] Label data: DN severity classification (mild, moderate, severe) and continuous glucose monitoring (CGM) data

[0199] Training Process: Divide the dataset into training set, validation set and test set, and use cross-validation to optimize model hyperparameters.

[0200] Train the model using the backpropagation algorithm and Adam optimizer, with optional loss functions:

[0201] Cross-Entropy Loss: Suitable for classification tasks (e.g., hypoglycemic risk grading).

[0202] Mean Squared Error: Suitable for regression tasks (e.g., DN progression probability prediction).

[0203] Monitor validation set performance to prevent overfitting.

[0204] Model evaluation: Evaluate the model's prediction accuracy, sensitivity, and specificity on the test set to ensure generalization capability.

[0205] On the test set, DN prediction uses F1-score (considering class imbalance), and hypoglycemic warning uses Mean Absolute Error (MAE).

[0206] Dynamic threshold performance: Evaluate different threshold strategies through the Area Under the Receiver Operating Characteristic Curve (AUC), requiring at least a 5% improvement in personalized threshold AUC compared to static thresholds.

[0207] Model validation test passes, then it can be deployed on the background system for DN warning prediction.

[0208] 4. Risk prediction

[0209] Real-time application: The core algorithm layer receives real-time collected and processed biosignal data.

[0210] Forward propagation: Input data into the trained model, output:

[0211] DN progression probability: A value between 0 and 1, representing the probability of DN progression to a severe stage within the next 3 / 6 / 12 months.

[0212] Hypoglycemic risk index: Grading value (e.g., 0: low risk, 1: medium risk, 2: high risk) based on current signal features and historical trends.

[0213] Output application:

[0214] When the DN progression probability exceeds 0.5, recommend more frequent clinical examinations.

[0215] When the hypoglycemic risk index is 2 (high risk), the system triggers an alarm.

[0216] 5. Model updating and optimization

[0217] Online learning mechanism: Dynamically update the model based on individual patient data to adapt to physiological changes.

[0218] Update strategy:

[0219] Periodically (e.g., monthly) fine-tune the model using recent patient data and clinical feedback.

[0220] Adopt transfer learning, only update part of the layers to maintain computational efficiency.

[0221] Threshold adjustment: combine with dynamic threshold module, adjust the alarm triggering threshold according to the prediction results and patient historical data.

[0222] The early warning system can be responsible for processing multi-modal biosignal data obtained from the signal acquisition layer and the preprocessing engine, and predicting the risk of diabetic neuropathy (DN) progression and the risk of unconscious hypoglycemia using deep learning models.

[0223] Preferably, in step S4, according to the risk prediction results, it is judged whether to trigger an alarm, and the corresponding judgment results are output, including:

[0224] According to the individual data of DN patients, set the early warning threshold of DN patients on the DN progression probability and the hypoglycemia risk index respectively, including:

[0225] Based on the difference in HRV / BPV / ECG-v distribution between healthy people and DN patients, the basic threshold is determined by kernel density estimation;

[0226] Through the individual historical data of DN patients: the trend of HRV / BPV or the abnormal frequency of ECG-v, the threshold is dynamically adjusted by Bayesian optimization:

[0227] ,

[0228] Wherein:

[0229] is the basic threshold;

[0230] is the current threshold;

[0231] is the optimized threshold;

[0232] η is the set learning rate;

[0233] is the set log-likelihood gradient in the Bayesian function;

[0234] Obtain the risk prediction results of the DN patient, including the DN progression probability and the hypoglycemia risk index of the DN patient;

[0235] Judge whether the DN progression probability of the DN patient exceeds the early warning threshold of the DN patient on the DN progression probability:

[0236] If yes, generate the corresponding probability of DN clinical examination strategy; otherwise, give up;

[0237] determining whether the hypoglycemia risk index of the DN patient exceeds a warning threshold for the hypoglycemia risk index of the DN patient:

[0238] if yes, triggering an alarm and generating an alarm signal for the corresponding DN patient; otherwise, giving up.

[0239] Because each DN patient has individual differences, different warning thresholds need to be set for them for individualized warning. A dynamic threshold module is provided in the background, which can dynamically adjust the warning threshold of each DN patient based on individual data of the DN patient, including the warning thresholds on the DN progression probability and the hypoglycemia risk index, respectively.

[0240] Based on the differences in HRV / BPV / ECG-v distribution between healthy people and DN patients, a kernel density estimation is used to determine the basic threshold, for example, taking the difference in HRV distribution as an example:

[0241] Based on the HRV parameters (such as SDNN, RMSSD, and LF / HF ratio) of healthy people and DN patients, a non-parametric probability density function is constructed:

[0242] ,

[0243] is the HRV sample data of the DN patient, and x is the HRV sample data of the healthy population (such as n≥1000 healthy people and n≥300 DN patients);

[0244] is a Gaussian kernel function, and the bandwidth h is optimized by the Silverman criterion;

[0245] Basic threshold setting principle: take the 95th percentile of the healthy population distribution as the initial abnormal threshold (such as SDNN<50ms triggering a warning).

[0246] The steps of Bayesian optimization are as follows: please understand the mapping relationship between the threshold parameter and the warning performance by combining the Bayesian optimization framework: Gaussian Process Regression (GP).

[0247] By fusing the group distribution characteristics and individual dynamic data, the adaptability defects of traditional fixed threshold models in diabetic neuropathy monitoring are overcome, especially suitable for dynamic tracking of HRV nonlinear attenuation (such as LF / HF ratio mutation) and ECG repolarization abnormalities caused by autonomic nerve remodeling.

[0248] When the system is applied:

[0249] To activate the smart wearable device for the user, the device collects continuous data and uploads to the background;

[0250] The pre-processing engine extracts features (e.g. HRV multiscale entropy, ECG morphological parameters) in real time;

[0251] The feature data is input into the trained LSTM-CNN model (DN early warning model) to generate risk prediction results;

[0252] The dynamic threshold module evaluates the risk index to determine whether to trigger an alarm (e.g. vibration prompt or push notification to PDA / EDA care terminal);

[0253] The system stores the results and uploads them to the cloud regularly for medical staff analysis (administrator background view).

[0254] Figure 3 is a block diagram of a dynamic early warning device for diabetic neuropathy based on multi-modal biological signals according to an exemplary embodiment, which is applied to a dynamic early warning method for diabetic neuropathy based on multi-modal biological signals. Referring to Figure 3 , on the other hand, a dynamic early warning device for diabetic neuropathy based on multi-modal biological signals is provided, which is applied to a dynamic early warning method for diabetic neuropathy based on multi-modal biological signals, and the device comprises:

[0255] The smart wearable device is used to collect biological signal data of DN patients in real time: heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG), and input into the pre-set pre-processing engine;

[0256] The pre-processing engine is used to pre-process the biological signal data, and extract and obtain feature data of DN patients in real time: heart rate variability parameters (HRV), blood pressure variability parameters (BPV) and electrocardiogram morphological parameters (ECG-v), and input the feature data into the early warning system;

[0257] The early warning system is used to input the feature data into the pre-deployed DN dynamic early warning model, which identifies the feature data and outputs corresponding DN risk prediction results, including DN progression probability and hypoglycemia risk index;

[0258] The dynamic threshold module is used to determine whether the risk prediction results trigger an alarm according to the pre-set early warning threshold of DN patients in DN progression probability and hypoglycemia risk index, and output the corresponding judgment results;

[0259] The data service module is used to record the biological signal data, feature data, risk prediction results and judgment results of DN patients, and store and upload them to the cloud.

[0260] Figure 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present invention, such as Figure 4 As shown, the electronic device may include the above Figure 3 Optionally, the electronic device 410 may include a first processor 2001 .

[0261] Optionally, the electronic device 410 may further include a memory 2002 and a transceiver 2003 .

[0262] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0263] The following combination Figure 4 The components of the electronic device 410 are described in detail.

[0264] The first processor 2001 is the control center of the electronic device 410 and can be a single processor or a collective term for multiple processing elements. For example, the first processor 2001 can be one or more central processing units (CPUs), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention, such as one or more digital signal processors (DSPs) or one or more field programmable gate arrays (FPGAs).

[0265] Optionally, the first processor 2001 can execute various functions of the electronic device 410 by running or executing software programs stored in the memory 2002 and calling data stored in the memory 2002.

[0266] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Figure 4 CPU0 and CPU1 are shown in FIG.

[0267] In a specific implementation, as an embodiment, the electronic device 410 may also include multiple processors, such as Figure 4The first processor 2001 and the second processor 2004 shown in the foregoing embodiments can be implemented by using a single-CPU or a multi-CPU. The processor herein can refer to one or more devices, circuits, and / or processing cores for processing data (for example, computer program instructions).

[0268] The memory 2002 is configured to store a software program for implementing the scheme of the present application, and the first processor 2001 is configured to control the execution of the software program. For details, refer to the method embodiments described above, which will not be repeated here.

[0269] Optionally, the memory 2002 can be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disk storage, a magneto-optical disk, a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer, but is not limited to this. The memory 2002 can be integrated with the first processor 2001 or exist independently and be coupled to the first processor 2001 through an interface circuit (not shown in the foregoing embodiments) of the electronic device 410. The embodiments of the present application do not make a specific limitation in this regard. Figure 4

[0270] The transceiver 2003 is configured to communicate with a network device or a terminal device.

[0271] Optionally, the transceiver 2003 can include a receiver and a transmitter (not shown in the foregoing embodiments). Figure 4 The receiver is configured to implement the receiving function, and the transmitter is configured to implement the transmitting function.

[0272] Optionally, the transceiver 2003 can be integrated with the first processor 2001 or exist independently and be coupled to the first processor 2001 through an interface circuit (not shown in the foregoing embodiments) of the electronic device 410. The embodiments of the present application do not make a specific limitation in this regard. Figure 4

[0273] It should be noted that,​​​ The structure of the electronic device 410 shown in the figure does not constitute a limitation on the router, and an actual knowledge structure identification device can include more or fewer components than the figure, or combine certain components, or different component arrangements.

[0274] In addition, the technical effects of the electronic device 410 can refer to the technical effects of the diabetes neuropathy dynamic early warning method based on the multi-modal biological signal described in the above method embodiments, which will not be described here.

[0275] It should be understood that the first processor 2001 in the embodiments of the present application can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), ready programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0276] It should also be understood that the memory in the embodiments of the present application can be volatile or nonvolatile memory, or can include both volatile and nonvolatile memory. The nonvolatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable PROM (EPROM), electrically EPROM (EEPROM), or flash memory. The volatile memory can be random access memory (RAM) used as external cache. By way of example, and not limitation, many forms of random access memory (RAM) are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0277] The above-described embodiments can be implemented in whole or in part by software, hardware (such as a circuit), firmware, or any combination thereof. When implemented in software, the above-described embodiments can be implemented in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0278] It should be understood that the term "and / or" herein merely describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents that the associated objects before and after it are in an "or" relationship, but it can also represent an "and / or" relationship, which can be understood according to the context before and after it.

[0279] In the present application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or the like means any combination of the items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0280] It should be understood that in various embodiments of the present application, the size of the sequence number of the above-described processes does not mean the order of execution, and the execution order of the processes should be determined according to their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0281] Those skilled in the art can clearly understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are performed 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 the present application.

[0282] Those skilled in the art can clearly understand that, for the convenience and brevity of the description, the specific working processes of the devices, apparatuses and units described above can refer to the corresponding processes in the foregoing method embodiments, which will not be repeated here.

[0283] In several embodiments provided by the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0284] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0285] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit.

[0286] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the parts of the present application that essentially contribute to the prior art or the parts of the technical solutions can be embodied in the form of software products. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0287] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A dynamic early warning method for diabetic neuropathy based on multimodal biosignals. This method integrates multimodal sensors into smart wearable devices to collect heart rate (HR), blood pressure (BP), and single-lead electrocardiogram (ECG) signals. A personalized prediction model is constructed using a deep learning algorithm. This method uses artificial intelligence algorithms to predict the risk of diabetic neuropathy progression and identify the risk of unaware hypoglycemia in real time, triggering graded intervention measures. The method is characterized by: The method comprises: S1. Real-time collection of biosignal data from DN patients: heart rate (HR), blood pressure (BP), and single-lead electrocardiogram (ECG), and input into a preset pre-processing engine; S2. The preprocessing engine preprocesses the biosignal data and extracts and obtains characteristic data of DN patients in real time: heart rate variability parameter (HRV), blood pressure variability parameter (BPV) and electrocardiogram morphology parameter (ECG-v); S3. Inputting the characteristic data into a pre-deployed DN dynamic early warning model, which recognizes the characteristic data and outputs corresponding DN risk prediction results: DN progression probability and hypoglycemia risk index. The method for generating the DN early warning model includes the following steps: Acquire biosignal datasets from several DN patients, including multimodal biosignals at different time series: heart rate (HR), blood pressure (BP), and single-lead electrocardiogram (ECG), as well as labeled data: DN severity classification and continuous glucose monitoring (CGM) data; Preprocessing the biosignal data of each DN patient through the preprocessing engine and extracting corresponding feature data in real time: heart rate variability parameter (HRV), blood pressure variability parameter (BPV) and electrocardiogram morphology parameter (ECG-v); According to the DN severity grade, the characteristic data is marked with a corresponding DN progression probability label; According to the continuous glucose monitoring (CGM) data, the characteristic data is labeled with a corresponding hypoglycemia risk index; The characteristic data of each DN patient are collected to obtain a data set for training the model, and the data set is divided into a training set, a validation set, and a test set according to the proportion; Inputting the training set into a preset LSTM-CNN model for feature learning, and training to generate the initial DN warning model; The application performance of the DN early warning model was verified using the validation set: If the verification is successful, the prediction performance of the DN early warning model is tested using the test set. If successful, the DN early warning model is deployed and applied. Otherwise, retraining is performed. Otherwise, retrain; S4. Determine whether to trigger an alarm based on the risk prediction result, and output a corresponding determination result; S5. Record the biological signal data, characteristic data, risk prediction results and judgment results of DN patients, store them and upload them to the cloud.

2. The dynamic early warning method for diabetic neuropathy based on multimodal biosignals according to claim 1, characterized in that: In step S1, the method of collecting biological signal data of DN patients in real time is as follows: The biosignal data including heart rate (HR), blood pressure (BP) and single-lead electrocardiogram (ECG) are collected through smart wearable devices worn by DN patients; as well as, The biosignal data is subjected to motion artifact correction using a motion artifact correction algorithm, and the corrected data is uploaded to the preprocessing engine.

3. The dynamic early warning method for diabetic neuropathy based on multimodal biosignals according to claim 1, characterized in that: In step S4, based on the risk prediction result, it is determined whether an alarm is triggered and a corresponding determination result is output, including: Based on the individual data of DN patients, early warning thresholds for DN progression probability and hypoglycemia risk index are set in advance, including: Based on the differences in HRV / BPV / ECG-v distribution between healthy subjects and DN patients, kernel density estimation was used to determine the basic threshold; The threshold is dynamically adjusted using Bayesian optimization based on the individual historical data of DN patients: HRV / BPV trends or ECG-v abnormality frequencies: ， in: is the basic threshold; is the current threshold; is the optimized threshold; η is the set learning rate; is the log-likelihood gradient set in the Bayesian function; Obtaining the risk prediction results of the DN patient, including the DN progression probability and hypoglycemia risk index of the DN patient; Determine whether the probability of DN progression in DN patients exceeds the warning threshold of DN progression probability: If yes, generate a DN clinical examination strategy with the corresponding probability; otherwise, give up; Determine whether the hypoglycemia risk index of DN patients exceeds the warning threshold of the hypoglycemia risk index of DN patients: If so, an alarm is triggered and an alarm signal for the corresponding DN patient is generated; otherwise, the alarm is abandoned.

4. A dynamic early warning device for diabetic neuropathy based on multimodal biosignals, wherein the dynamic early warning device for diabetic neuropathy based on multimodal biosignals is used to implement the dynamic early warning method for diabetic neuropathy based on multimodal biosignals as described in any one of claims 1 to 3, characterized in that: The device comprises: Smart wearable devices are used to collect real-time biosignal data from DN patients: heart rate (HR), blood pressure (BP), and single-lead electrocardiogram (ECG), and input them into a preset pre-processing engine; A preprocessing engine, configured to preprocess the biosignal data, extract and obtain characteristic data of DN patients in real time: heart rate variability parameters (HRV), blood pressure variability parameters (BPV), and electrocardiogram morphology parameters (ECG-v), and input the characteristic data into an early warning system; An early warning system is used to input the characteristic data into a pre-deployed DN dynamic early warning model, which recognizes the characteristic data and outputs corresponding DN risk prediction results, including DN progression probability and hypoglycemia risk index; A dynamic threshold module is used to determine whether the risk prediction result triggers an alarm based on the pre-set warning thresholds of DN progression probability and hypoglycemia risk index for DN patients, and output the corresponding judgment result; The data service module is used to record the biological signal data, characteristic data, risk prediction results and judgment results of DN patients, and store and upload them to the cloud.

5. An electronic device, characterized in that: The electronic device comprises: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method according to any one of claims 1 to 3 is implemented.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program code, which can be called by a processor to execute the method according to any one of claims 1 to 3.

Images