Portable dynamic high-frequency electrocardiogram monitoring method and system

Abstract

The invention provides a portable dynamic high-frequency electrocardio monitoring system and method, and provides an efficient solution aiming at the problems that traditional high-frequency electrocardio depends on exercise load, diagnosis is delayed, the detection rate is low and the like in myocardial ischemia diagnosis. Electrocardio acquisition hardware of the system adopts a high signal-to-noise ratio and high-fidelity design so as to enhance the detection capability of high-frequency electrocardiosignals, and data is transmitted to a mobile terminal in real time through a wireless module, so that continuous dynamic monitoring is realized. A software algorithm preprocesses the collected dynamic resting high-frequency electrocardio data, abnormal waveforms are removed, and high-frequency signals are extracted with 3 seconds as a unit. Then, the data are input into a pre-trained electrocardiogram classification network, the network integrates a multi-scale convolution module, a feature splicing module, a long-short term memory network module, a self-adaptive convolution module and other modules, automatic feature extraction and time sequence modeling are achieved, and early myocardial ischemia can be efficiently recognized. The system provides an efficient and real-time solution for early diagnosis and intervention of myocardial ischemia.

Description

Technical Field

[0001] The present invention relates to the technical field of medical instruments, and particularly to a portable dynamic high-frequency electrocardiogram monitoring method and system. Background Art

[0002] Cardiovascular diseases are the biggest threat to humans. As a non-invasive and rapid diagnostic method, electrocardiogram has become the most routine detection method for cardiovascular diseases. High-frequency electrocardiogram is a technology developed on the basis of conventional electrocardiogram, mainly analyzing electrocardiogram signals in the range of 150-250 Hz. Existing research has shown that high-frequency electrocardiogram has a higher detection rate for coronary heart disease than conventional electrocardiogram. The conventional high-frequency electrocardiogram monitoring method relies on extracting indicators such as RMS for analysis, and it is more diagnostically valuable when the patient's heart rate reaches more than 80% of the maximum heart rate corresponding to their age under exercise load. This monitoring method has a relatively long acquisition time, about more than 6 minutes. And during the acquisition process, if situations such as the heart rate not reaching the threshold, excessive noise, or too many noise leads occur, this analysis method will fail, making the collected data of the patient lose diagnostic significance. In addition, many patients may not be able to reach the heart rate threshold through exercise, and this acquisition method is very unfriendly to this part of patients. First, collect high-frequency electrocardiogram information during exercise, select the high-frequency QRS waveform curve within a preset time period, select the first reference point with the minimum root mean square voltage on the high-frequency QRS reference waveform curve, and set the first screening condition and the second screening condition to obtain the second reference point and the third reference point. Determine the relative value of amplitude decrease according to the first reference point and the second reference point. If the relative value of amplitude decrease and the relative value of amplitude increase meet the preset conditions, determine the reference information according to the high-frequency QRS waveform curve. This method uses high-frequency electrocardiogram during exercise load to diagnose myocardial ischemia and myocardial infarction.

[0003] The prior art relies on the acquisition of high-frequency electrocardiogram during exercise load, and requires the patient to record the electrocardiogram during exercise in order to induce changes in the electrocardiogram signal for diagnosis. This has certain limitations for some patients with difficulty in exercising (such as the elderly, seriously ill patients, etc.). According to the electrocardiogram record during exercise in the prior art, there will be a lag in analysis, and data acquisition and analysis are carried out only after the patient exercises. This means that the patient needs to be tested under specific conditions, and the analysis is usually carried out offline and cannot provide real-time feedback. Summary of the Invention

[0004] By providing a portable dynamic high-frequency electrocardiogram monitoring method, the present invention solves the problem in the prior art that after the patient undergoes exercise load training, the electrocardiogram is recorded and then diagnosed, improving the diagnostic efficiency and accuracy.

[0005] In a first aspect, the present invention provides a portable dynamic high-frequency electrocardiogram monitoring method, which includes:

[0006] Obtain resting electrocardiogram data, perform data preprocessing on the resting electrocardiogram data to obtain resting high-frequency electrocardiogram data;

[0007] Input the resting high-frequency electrocardiogram data into a pre-trained electrocardiogram signal classification network to obtain the monitoring result of the resting high-frequency electrocardiogram for myocardial ischemia; wherein, the electrocardiogram signal classification network includes: a multi-scale convolutional neural module, a feature splicing module, a long short-term memory module, an adaptive convolutional module, and a classification module;

[0008] The multi-scale convolutional neural module is used to perform multi-scale feature extraction on the resting high-frequency electrocardiogram data to obtain local detail features corresponding to each convolutional layer;

[0009] The feature splicing module is used to splice the local detail features corresponding to each convolutional layer to obtain an activation fusion splicing feature;

[0010] The long short-term memory module is used to perform temporal modeling on the activation fusion splicing feature to obtain temporal correlation information;

[0011] The adaptive convolutional module is used to randomly generate dynamic convolution kernels to perform feature extraction on the temporal correlation information to obtain temporal detail features corresponding to each convolutional layer;

[0012] The classification module is used to classify the temporal detail features corresponding to each convolutional layer to obtain a classification result.

[0013] Combined with the first aspect, in a possible implementation manner, the performing data preprocessing on the resting electrocardiogram data to obtain resting high-frequency electrocardiogram data includes:

[0014] Split the resting electrocardiogram data to obtain multiple groups of split resting electrocardiogram data;

[0015] After cleaning the multiple groups of split resting electrocardiogram data, calculate the correlation coefficient between the data within each group of split resting electrocardiogram data, and determine whether the correlation coefficient is less than a first preset value;

[0016] If the correlation coefficient between the data within a group of split resting electrocardiogram data is less than the first preset value, discard the split resting electrocardiogram data from the multiple groups of split resting electrocardiogram data;

[0017] For the split resting electrocardiogram data with a correlation coefficient between the data within the group not less than the first preset value, extract high-frequency electrocardiogram signals from the split resting electrocardiogram data through a band-pass filter to obtain resting high-frequency electrocardiogram data.

[0018] In a possible implementation manner in combination with the first aspect, the multi-scale convolutional neural module includes a first convolutional layer, a second convolutional layer, and a third convolutional layer arranged in parallel;

[0019] The multi-scale convolutional neural module is used to perform multi-scale feature extraction on the resting high-frequency electrocardiogram data to obtain local detail features corresponding to each convolutional layer, including:

[0020] Performing fast fluctuation detail feature extraction on the QRS waveband in the resting high-frequency electrocardiogram data according to the first convolutional layer to obtain a first feature;

[0021] Performing overall morphological feature extraction on the QRS waveband in the resting high-frequency electrocardiogram data according to the second convolutional layer to obtain a second feature;

[0022] Performing heart rate transformation and QRS intensity feature extraction on the QRS waveband in the resting high-frequency electrocardiogram data according to the third convolutional layer to obtain a third feature;

[0023] Taking the first feature, the second feature, and the third feature as the local detail features corresponding to each convolutional layer.

[0024] In a possible implementation manner in combination with the first aspect, the feature splicing module includes: a filter splicing layer, a batch normalization layer, and a ReLU activation function layer;

[0025] The feature splicing module is used to splice the local detail features corresponding to each branch to obtain an activation fusion splicing feature, including:

[0026] Splicing the first feature, the second feature, and the third feature according to the filter splicing layer to obtain an initial activation fusion splicing feature;

[0027] Performing a normalization operation on the initial activation fusion splicing feature according to the batch normalization layer to obtain a normalized splicing feature;

[0028] Performing a non-linear operation on the normalized splicing feature according to the ReLU activation function layer to obtain an activation fusion splicing feature.

[0029] In a possible implementation manner in combination with the first aspect, the long short-term memory module includes a TimeDistributed layer and an LSTM layer connected in sequence;

[0030] The long short-term memory module is used to perform temporal modeling on the activation fusion splicing feature to obtain temporal correlation information, including:

[0031] Flatten the activated fusion concatenated features at each time step according to the TimeDistributed layer to obtain a one-dimensional feature vector corresponding to each time step;

[0032] Use the LSTM layer to perform temporal modeling on the one-dimensional feature vector corresponding to each time step, and capture long-term and short-term dependencies to obtain temporally relevant information.

[0033] Combined with the first aspect, in a possible implementation, the adaptive convolution module includes a fourth convolutional layer, a fifth convolutional layer, and a sixth convolutional layer in parallel; wherein, the number of convolutional kernels of the convolutional layers in the fourth convolutional layer, the fifth convolutional layer, and the sixth convolutional layer adaptively changes according to the temporally relevant information;

[0034] The adaptive convolution module is used to randomly generate dynamic convolution kernels to extract features from the temporally relevant information, and obtain temporally detailed features corresponding to each convolutional layer, including:

[0035] Extract the small-scale high-frequency transient features of the temporally relevant information according to the fourth convolutional layer to obtain the first self-convolutional temporally relevant information;

[0036] Extract the medium-scale intermediate-frequency morphological features of the temporally relevant information according to the fifth convolutional layer to obtain the second self-convolutional temporally relevant information;

[0037] Extract the large-scale low-frequency rhythm features of the temporally relevant information according to the sixth convolutional layer to obtain the third self-convolutional temporally relevant information.

[0038] Combined with the first aspect, in a possible implementation, the classification module includes: a channel feature concatenation layer, a batch normalization activation flattening layer, a regularization fully connected activation layer, and a fully connected discrimination layer;

[0039] The classification module is used to classify the temporally detailed features corresponding to each convolutional layer to obtain a classification result, including:

[0040] According to the channel feature concatenation layer, concatenate the temporally detailed features corresponding to each convolutional layer to obtain the initial concatenated self-convolutional temporally relevant information;

[0041] According to the batch normalization activation flattening layer, perform batch normalization, activation, and flattening operations on the initial concatenated self-convolutional temporally relevant information to obtain one-dimensional self-convolutional temporal features;

[0042] According to the regularization fully connected activation layer, perform regularization and compression operations on the one-dimensional self-convolutional temporal features to obtain compressed self-convolutional temporal features;

[0043] According to the fully-connected discriminant layer, classify the compressed self-convolution time series features to obtain a classification result.

[0044] Combined with the first aspect, in a possible implementation manner, the loss function used in the process of training the electrocardiogram signal classification network is:

[0045]

[0046] where N represents the total number of training samples; y i represents the true label of the i-th training sample; p i represents the prediction result of the i-th training sample.

[0047] In a second aspect, the present invention provides a portable dynamic high-frequency electrocardiogram monitoring system, which includes: a front-end acquisition and processing module, a signal primary processing module, a wireless communication module, a result prediction module, and a power supply module;

[0048] The front-end acquisition and processing module: is used to acquire multi-channel body surface potentials, and obtain primary high-frequency electrocardiogram signals by performing differential amplification and AD conversion on the high-frequency electrocardiogram signals;

[0049] The signal primary processing module: is used to perform primary processing on the acquired primary high-frequency electrocardiogram signals, and obtain a denoised standard high-frequency electrocardiogram signal through primary processing;

[0050] The wireless communication module: is used to transmit the standard high-frequency electrocardiogram signal to a corresponding display module according to the wireless transmission function of the integrated Bluetooth module;

[0051] The result prediction module: is used to perform result prediction according to a pre-trained electrocardiogram signal classification network and the preprocessed high-frequency electrocardiogram signal to obtain a monitoring result of the resting high-frequency electrocardiogram for myocardial ischemia; wherein, the pre-trained electrocardiogram signal classification network is deployed on the dynamic high-frequency electrocardiogram monitoring system;

[0052] The power supply module: is used to supply power to the front-end acquisition and processing module, the wireless communication module, and the signal primary processing module.

[0053] Combined with the second aspect, in a possible implementation manner, the front-end acquisition and processing module includes: the front-end acquisition and processing module includes: an acquisition unit, a filtering unit, a buffering unit, a right leg drive noise reduction unit, a differential amplification unit, and an AD conversion unit;

[0054] The front-end acquisition and processing module: is used to acquire multi-channel body surface potentials multiple times, and obtain multiple groups of primary high-frequency electrocardiogram signals by performing differential amplification and AD conversion on the high-frequency electrocardiogram signals, including:

[0055] The acquisition unit: is used to collect multiple groups of multi-channel body surface potentials by using an electrocardiogram sensor; wherein, each group of multi-channel body surface potentials includes: body surface potentials at electrode positions LA, LL, RA, and V1-V6;

[0056] The filtering unit: is used to perform low-pass filtering on the multi-channel body surface potentials to obtain the low-pass filtered multi-channel body surface potentials;

[0057] The buffer unit: is used to provide a low output impedance for the low-pass filtered multi-channel body surface potentials to obtain the buffered multi-channel body surface potentials;

[0058] The right leg drive noise reduction unit: is used to perform noise reduction processing on the buffered multi-channel body surface potentials to obtain the noise-reduced multi-channel body surface potentials;

[0059] The differential amplification unit: is used to perform differential processing on the noise-reduced multi-channel body surface potentials to obtain multiple groups of differential amplification signals;

[0060] The AD conversion unit: is used to perform AD conversion on the multiple groups of differential amplification signals to obtain multiple groups of primary high-frequency electrocardiogram signals.

[0061] One or more technical solutions provided in the present invention have at least the following technical effects or advantages: By acquiring resting electrocardiogram data and performing data preprocessing on the resting electrocardiogram data, resting high-frequency electrocardiogram data is obtained; The present invention uses the resting high-frequency electrocardiogram data, does not require patients to perform exercise load training, and is applicable to more people; Inputting the resting high-frequency electrocardiogram data into a pre-trained electrocardiogram signal classification network to obtain the monitoring results of the resting high-frequency electrocardiogram for myocardial ischemia; wherein, the electrocardiogram signal classification network includes: a multi-scale convolutional neural module, a feature splicing module, a long short-term memory module, an adaptive convolutional module, and a classification module; The multi-scale convolutional neural module is used to perform multi-branch multi-scale feature extraction on the resting high-frequency electrocardiogram data to obtain local detail features corresponding to each branch; Performing multi-scale feature extraction on the input resting high-frequency electrocardiogram data can capture the local details of the resting high-frequency electrocardiogram data and avoid the limitations of traditional manual feature extraction; The long short-term memory module is used to perform temporal modeling on the local detail features corresponding to each convolution to obtain temporal encoding and hidden layer features; The LSTM module simultaneously focuses on the temporal relationship before and after the local detail features corresponding to each convolution to obtain temporally relevant information, and has a stronger recognition ability for rapidly changing pathological signals in the resting high-frequency electrocardiogram data; The adaptive convolutional module is used to randomly generate dynamic convolution kernels to extract features from the temporally relevant information to obtain temporal detail features corresponding to each convolutional layer; By highlighting the attention to key electrocardiogram segments through the temporal detail features, the ability to capture pathological events such as myocardial ischemia and myocardial infarction is enhanced. Description of the Drawings

[0062] Figure 1 This is the flowchart of the steps of the portable dynamic high-frequency electrocardiogram monitoring method provided by the embodiments of the present invention;

[0063] Figure 2 This is the schematic diagram of the electrocardiogram signal classification network structure provided by the embodiments of the present invention;

[0064] Figure 3 This is the schematic diagram of the electrode placement of the bipolar limb leads and the human limbs provided by the embodiments of the present invention;

[0065] Figure 4 This is the schematic diagram of the portable dynamic high-frequency electrocardiogram monitoring system provided by the embodiments of the present invention. Specific embodiments

[0066] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0067] In the first aspect, the present invention provides a portable dynamic high-frequency electrocardiogram monitoring method. Refer to Figure 1 , this method includes the following steps S101 to S102.

[0068] S101, obtain resting electrocardiogram data, perform data preprocessing on the resting electrocardiogram data to obtain resting high-frequency electrocardiogram data;

[0069] Specifically, in step S101, obtaining resting electrocardiogram data and performing data preprocessing on the resting electrocardiogram data to obtain resting high-frequency electrocardiogram data includes the following steps S1011 to S1012.

[0070] S1011, split the resting electrocardiogram data to obtain multiple groups of split resting electrocardiogram data;

[0071] S1012, after cleaning the multiple groups of split resting electrocardiogram data, calculate the correlation coefficient between the intra-group data of each group of split resting electrocardiogram data, and determine whether the correlation coefficient is less than a first preset value;

[0072] If the correlation coefficient between the intra-group data of the split resting electrocardiogram data is less than the first preset value, discard the split resting electrocardiogram data from the multiple groups of split resting electrocardiogram data;

[0073] For the split resting electrocardiogram data with the correlation coefficient between the intra-group data not less than the first preset value, the high-frequency electrocardiogram signal is extracted from the split resting electrocardiogram data through a band-pass filter to obtain the resting high-frequency electrocardiogram data.

[0074] S102, input the resting high-frequency electrocardiogram data into the pre-trained electrocardiogram signal classification network to obtain the monitoring result of the resting high-frequency electrocardiogram for myocardial ischemia; see Figure 2 , where the electrocardiogram signal classification network includes: a multi-scale convolutional neural module, a feature splicing module, a long short-term memory module, an adaptive convolutional module, and a classification module.

[0075] The multi-scale convolutional neural module is used to perform multi-scale feature extraction on the resting high-frequency electrocardiogram data to obtain the local detail features corresponding to each convolutional layer.

[0076] Specifically, the multi-scale convolutional neural module includes a first convolutional layer, a second convolutional layer, and a third convolutional layer in parallel;

[0077] The multi-scale convolutional neural module is used to perform multi-scale feature extraction on the resting high-frequency electrocardiogram data to obtain the local detail features corresponding to each convolutional layer, including:

[0078] According to the first convolutional layer, the rapid fluctuation detail features of the QRS band in the resting high-frequency electrocardiogram data are extracted to obtain the first feature;

[0079] According to the second convolutional layer, the overall morphological features of the QRS band in the resting high-frequency electrocardiogram data are extracted to obtain the second feature;

[0080] According to the third convolutional layer, the heart rate transformation and QRS intensity features of the QRS band in the resting high-frequency electrocardiogram data are extracted to obtain the third feature;

[0081] The first feature, the second feature, and the third feature are used as the local detail features corresponding to each convolutional layer.

[0082] It can be understood that the first convolutional layer uses a 1*31 convolutional kernel to extract features from the resting high-frequency electrocardiogram data, focusing on capturing the rapid fluctuation details of the QRS band, mainly extracting the steep edges and rapidly changing features in the QRS band to obtain the first feature.

[0083] The second convolutional layer is used to extract the overall morphological features of the QRS band in the resting high-frequency electrocardiogram data using a 1*36 convolutional kernel to obtain the second feature;

[0084] It can be understood that the second convolutional layer uses a 1*36 convolutional kernel to extract features from the resting high-frequency electrocardiogram data, focusing on capturing the overall morphological features of the QRS complex, extracting features with a certain time dependence, to obtain the second feature.

[0085] The third convolutional layer is used to perform heart rate transformation and QRS intensity feature extraction on the QRS band in the resting high-frequency electrocardiogram data by using a 1*41 convolutional kernel, and obtain the third feature;

[0086] It can be understood that the third convolutional layer uses a 1*41 convolutional kernel to extract features from the resting high-frequency electrocardiogram data, focuses on capturing the rhythm and trend information of the QRS complex, extracts features including heart rate transformation and QRS intensity, and obtains the third feature.

[0087] Exemplarily, the multi-scale convolutional neural module is based on an input size of 12x3000, and adopts a three-layer one-dimensional convolutional (Conv1D) structure, which respectively includes 31, 36, and 41 filters, the convolutional kernel size is 3, and the stride is 5. The output dimension of the last layer Conv1D-41 is 3x3x600, indicating that local high-frequency features are extracted through multi-channel convolution, and the receptive field is gradually expanded by stacking layer by layer to provide a hierarchical time-domain feature expression for the subsequent module. The addition of Dropout and BatchNormalization further enhances the generalization ability of the model. Information at different time scales in the high-frequency electrocardiogram signal is extracted through convolutional kernels of different sizes, and the features extracted at each scale are spliced into a fused feature vector. Subsequently, batch normalization is performed on the fused feature vector to stabilize the input of each layer, accelerate network convergence, and introduce non-linearity through the ReLU activation function to enhance the feature expression ability.

[0088] The feature splicing module is used to splice the local detail features corresponding to each convolutional layer to obtain the activated fusion spliced feature;

[0089] Specifically, the feature splicing module includes: a filter splicing layer, a batch normalization layer, and a ReLU activation function layer;

[0090] The feature splicing module is used to splice the local detail features corresponding to each branch to obtain the activated fusion spliced feature, including:

[0091] Splice the first feature, the second feature, and the third feature according to the filter splicing layer to obtain the initial activated fusion spliced feature;

[0092] Perform a normalization operation on the initial activated fusion spliced feature according to the batch normalization layer to obtain the normalized spliced feature;

[0093] Perform a non-linear operation on the normalized spliced feature according to the ReLU activation function layer to obtain the activated fusion spliced feature.

[0094] Exemplarily, the local detail features corresponding to each convolutional layer are concatenated to form a 9×600 fused feature matrix, and further batch normalization (BatchNormalization) and ReLU activation are performed to obtain the activated fused concatenated features, so as to ensure the stability of the features, accelerate the model convergence speed, and improve the feature expression ability at the same time.

[0095] The long short-term memory module is used to perform temporal modeling on the activated fused concatenated features to obtain temporal related information.

[0096] Specifically, it includes a TimeDistributed layer and an LSTM layer connected in sequence.

[0097] The long short-term memory module is used to perform temporal modeling on the activated fused concatenated features to obtain temporal related information, including:

[0098] Flatten the activated fused concatenated features at each time step according to the TimeDistributed layer to obtain a one-dimensional feature vector corresponding to each time step.

[0099] Use the LSTM layer to perform temporal modeling on the one-dimensional feature vector corresponding to each time step, and capture long-term and short-term dependencies to obtain temporal related information.

[0100] Exemplarily, in the temporal modeling stage, the LSTM layer processes the flattened temporal features. The LSTM layer has 200 hidden units, which are used to capture the temporal dependencies of the electrocardiogram signals, extract global time features, enable the network to recognize long-term signal pattern changes, and improve the perception ability of dynamic features. Capture the forward and backward dependencies of the signals and extract temporal related information. The LSTM layer performs in-depth modeling on the temporal features extracted by the multi-scale convolutional neural module through a long short-term memory module with 200 hidden units. Its core role is to capture the long-range dependencies of the electrocardiogram signals, such as the periodic features of arrhythmia or the temporal correlations of abnormal waveforms. The Transformation layer is combined with the feature dimension transformation to make the output of the LSTM layer match the input dimension of the adaptive convolution module Involution, forming an end-to-end joint optimization.

[0101] The adaptive convolution module is used to randomly generate dynamic convolution kernels to extract features from the temporal related information to obtain the temporal detail features corresponding to each convolutional layer.

[0102] Specifically, the adaptive convolution module includes a fourth convolutional layer, a fifth convolutional layer, and a sixth convolutional layer in parallel; among them, the number of convolution kernels of the convolutional layers in the fourth convolutional layer, the fifth convolutional layer, and the sixth convolutional layer changes adaptively according to the temporal related information.

[0103] An adaptive convolution module for randomly generating dynamic convolution kernels to extract feature information related to time series, obtaining time series detail features corresponding to each convolutional layer, including:

[0104] Extracting the small-scale high-frequency transient feature information related to time series according to the fourth convolutional layer to obtain the first self-convolution time series related information;

[0105] Extracting the medium-scale intermediate-frequency morphological feature information related to time series according to the fifth convolutional layer to obtain the second self-convolution time series related information;

[0106] Extracting the large-scale low-frequency rhythm feature information related to time series according to the sixth convolutional layer to obtain the third self-convolution time series related information.

[0107] Exemplarily, the network introduces the adaptive convolution module Involution. Different from traditional convolutions, this module can dynamically generate convolution kernels according to the input, thus more flexibly extracting local features. By setting different kernel_sizes, the Involution module further extracts features of different scales, splices and fuses them to obtain richer local spatio-temporal information. Involution adopts a channel-adaptive deconvolution structure. Involution-31 uses 5 channels and a stride of 4.5, while Involution-41 is compressed to 3 channels and outputs a feature map of 3x16x120. Compared with traditional convolutions, Involution enhances the ability to model the spatial-channel interaction of different lead signals while reducing the computational complexity by dynamically generating convolution kernel parameters, and is suitable for processing the multi-dimensional feature fusion of 12-lead electrocardiogram signals.

[0108] A classification module for classifying the time series detail features corresponding to each convolutional layer to obtain a classification result.

[0109] Specifically, the classification module includes: a channel feature splicing layer, a batch normalization activation flattening layer, a regularization fully connected activation layer, and a fully connected discrimination layer;

[0110] A classification module for classifying the time series detail features corresponding to each convolutional layer to obtain a classification result, including:

[0111] According to the channel feature splicing layer, splicing the time series detail features corresponding to each convolutional layer to obtain the initial spliced self-convolution time series related information;

[0112] According to the batch normalization activation flattening layer, performing batch normalization, activation, and flattening operations on the initial spliced self-convolution time series related information to obtain one-dimensional self-convolution time series features;

[0113] According to the regularization fully connected activation layer, performing regularization and compression operations on the one-dimensional self-convolution time series features to obtain compressed self-convolution time series features;

[0114] According to the fully connected discriminant layer, the compressed self-convolutional time series features are classified to obtain the classification result.

[0115] The loss function used in the process of training the electrocardiogram signal classification network is:

[0116]

[0117] where N represents the total number of training samples; y i represents the true label of the i-th training sample; p i represents the prediction result of the i-th training sample.

[0118] The present invention uses resting high-frequency electrocardiogram data for training and is completely independent of exercise load. After collecting a set number of electrocardiogram data, it can be directly input into the pre-trained electrocardiogram signal classification network to give the monitoring result of resting high-frequency electrocardiogram for myocardial ischemia in real time, eliminating the lag in the analysis process of traditional methods.

[0119] Through the automatic feature extraction of the electrocardiogram signal classification network, the present invention avoids the deficiencies of traditional methods that rely on manually designed feature extraction and classification algorithms, and improves the diagnostic efficiency and accuracy. After deep learning optimization, the diagnostic accuracy of electrocardiograms can reach more than 90%, significantly improving the early diagnostic ability of myocardial ischemia and myocardial infarction.

[0120] When training the electrocardiogram signal classification network: first perform abnormal electrocardiogram signal cleaning, and then perform signal filtering and noise removal.

[0121] Abnormal electrocardiogram signal cleaning: To build a high-quality electrocardiogram signal classification network, first train based on the resting high-frequency electrocardiogram data of 56 groups of patients. The acquisition duration of each group of data is about 3 minutes. First, extract the QRS waveforms in each group of high-frequency electrocardiogram data, and split each group of data into groups of 3000 points (i.e., 3 seconds).

[0122] To remove noise and abnormal waveforms, perform correlation coefficient analysis on each heartbeat rhythm within each group of 3000 points. If the correlation coefficient is less than 0.8, it is considered that the group of data contains abnormal waveforms or noise, and the group of data will be discarded. When performing QRS wave extraction and correlation coefficient analysis, synchronous processing of 12-lead signals is adopted. If the QRS waveform of a certain lead cannot be extracted, or the correlation coefficient is less than 0.8, all lead data in the group of data will be discarded. This process ensures the quality and consistency of the data used.

[0123] Signal filtering and noise removal: After performing the correlation coefficient analysis, we perform band-pass filtering on the data at 150 - 250 Hz to extract the resting high-frequency electrocardiogram data. Subsequently, the root mean square (RMS) value of the signal intensity outside the QRS complex is calculated as a measure of the noise level. If the noise level exceeds 1 μV, it indicates that the data in this group has excessive noise, and this group of data is discarded.

[0124] After the above data preprocessing, the resting high-frequency electrocardiogram data of 56 groups of patients are divided into 2022 groups of valid data, with each group having a data length of 3 seconds (i.e., 3000 data points). These data volumes are sufficient for deep learning training, and each group of data contains the label of whether myocardial ischemia exists annotated by doctors.

[0125] In a specific embodiment provided by the present invention, the deep learning adopted by the present invention is based on a multi-scale convolutional neural module and a long short-term memory module, and is better adapted to the multi-scale features and temporal dynamics of the resting high-frequency electrocardiogram data for the resting high-frequency electrocardiogram data.

[0126] Specifically, first in the multi-scale convolutional neural module, multi-scale feature extraction is performed on the input resting high-frequency electrocardiogram data; this process can capture the local detailed features of the QRS waveform in the resting high-frequency electrocardiogram data, avoiding the limitations of traditional manual feature extraction.

[0127] Subsequently, the multi-scale features are concatenated and the concatenated features are activated and fused for temporal modeling to obtain temporal correlation information. The LSTM layer simultaneously focuses on the temporal relationship before and after the local detailed features and has a stronger identification ability for the rapidly changing pathological signals in the resting high-frequency electrocardiogram data.

[0128] By acquiring resting electrocardiogram data and performing data preprocessing, the present invention obtains resting high-frequency electrocardiogram data, enabling it to be applicable to a wider range of people without the need for patients to undergo exercise stress tests. In terms of hardware, a portable high-frequency electrocardiogram acquisition system is designed, focusing on solving the technical problems of device portability and high-frequency signal acquisition. It adopts a multi-channel signal acquisition circuit, a filtering and input protection circuit, a buffer circuit, a differential amplification and AD conversion circuit, a right leg drive circuit, an electrocardiogram digital signal processor, and a wireless data transmission module. Each module collaborates efficiently to achieve high-quality electrocardiogram signal acquisition, real-time processing, and wireless transmission, ensuring that the device is lightweight and compact while having high-performance monitoring capabilities. In terms of software, the high-frequency electrocardiogram data processing flow is optimized. A multi-scale convolutional neural module is used to extract multi-scale features, and batch normalization and ReLU activation are combined to enhance the feature expression ability. Subsequently, the time series features are flattened through the TimeDistributed layer, and LSTM is used for time series modeling to extract the forward and backward dependencies of the signal. At the same time, Involution adaptive convolution is introduced to dynamically generate convolution kernels to extract local spatio-temporal features and combine different scale information fusion. Finally, the features are output for class prediction after being normalized, activated, flattened, and dimension-reduced by the fully connected layer, and the cross-entropy loss is used to optimize the model. While achieving high-quality electrocardiogram monitoring, the present invention greatly improves the portability of the device.

[0129] In a second aspect, the present invention provides a portable dynamic high-frequency electrocardiogram monitoring system for implementing a portable dynamic high-frequency electrocardiogram monitoring method. Refer to Figure 4 , the system includes: a front-end acquisition and processing module, a primary signal processing module, a wireless communication module, a result prediction module, and a power supply module;

[0130] The front-end acquisition and processing module: is used for collecting multi-channel body surface potentials multiple times, and obtaining multiple groups of primary high-frequency electrocardiogram signals through differential amplification and AD conversion of high-frequency electrocardiogram signals;

[0131] Here, multiple groups of primary high-frequency electrocardiogram signals refer to a set of multiple sampling data continuously acquired by the system at the original sampling frequency (16kSPS) after AD conversion. Each set represents a complete signal sampling captured within a very short period of time.

[0132] Here, the front-end acquisition and processing module includes: an acquisition unit, a filtering unit, a buffer unit, a right leg drive noise reduction unit, a differential amplification unit, and an AD conversion unit;

[0133] The front-end acquisition and processing module: is used for collecting multi-channel body surface potentials multiple times, and obtaining multiple groups of primary high-frequency electrocardiogram signals through differential amplification and AD conversion of high-frequency electrocardiogram signals, including:

[0134] Acquisition unit: used to collect multiple groups of multi-channel body surface potentials multiple times by means of an electrocardiogram sensor; among them, each group of multi-channel body surface potentials includes: body surface potentials at electrode positions LA, LL, RA, and V1 to V6;

[0135] Filtering unit: used to perform low-pass filtering on the multi-channel body surface potentials to obtain the low-pass filtered multi-channel body surface potentials;

[0136] Buffer unit: used to provide a low output impedance for the low-pass filtered multi-channel body surface potentials to obtain the buffered multi-channel body surface potentials;

[0137] Right leg drive noise reduction unit: used to perform noise reduction processing on the buffered multi-channel body surface potentials to obtain the noise-reduced multi-channel body surface potentials;

[0138] Differential amplification unit: used to perform differential processing on the noise-reduced multi-channel body surface potentials to obtain multiple groups of differential amplification signals;

[0139] AD conversion unit: used to perform AD conversion on multiple groups of differential amplification signals to obtain multiple groups of primary high-frequency electrocardiogram signals.

[0140] Signal primary processing module: used to perform summation downsampling on multiple groups of primary high-frequency electrocardiogram signals to obtain the noise-reduced primary high-frequency electrocardiogram signals; then perform calculations on the noise-reduced primary high-frequency electrocardiogram signals to obtain the standard high-frequency electrocardiogram signals;

[0141] Here, the signal primary processing module includes: a signal noise reduction module and a standard high-frequency electrocardiogram signal conversion module;

[0142] Signal noise reduction module: used to perform summation downsampling on multiple groups of data of the collected primary high-frequency electrocardiogram signals to obtain the noise-reduced primary high-frequency electrocardiogram signals;

[0143] Standard high-frequency electrocardiogram signal conversion module: used to convert the noise-reduced primary high-frequency electrocardiogram signals using the standard electrocardiogram signal conversion formula to obtain the standard high-frequency electrocardiogram signals, formulas (1.2) to (1.12).

[0144] Wireless communication module: used to transmit the standard high-frequency electrocardiogram signals to the corresponding display module according to the wireless transmission function of the integrated Bluetooth module;

[0145] Result prediction module: used to perform result prediction based on the pre-trained electrocardiogram signal classification network and the standard high-frequency electrocardiogram signals to obtain the monitoring results of resting high-frequency electrocardiogram for myocardial ischemia; among them, the pre-trained electrocardiogram signal classification network is deployed on the dynamic high-frequency electrocardiogram monitoring system;

[0146] Power supply module: used to supply power to the front-end acquisition and processing module, wireless communication module, and signal primary processing module.

[0147] Exemplarily, the principle of electrocardiogram (ECG) signal detection is essentially based on the weak periodic current generated by the heart during depolarization and repolarization. These currents are conducted throughout the body through human tissues and body fluids, causing periodic changes in the electric potential at various parts of the body surface. By placing electrodes at different positions on the human epidermis, corresponding potential signals can be collected, and the potential differences between different parts constitute the common electrocardiogram. As Figure 3 shown, the electrodes of bipolar limb leads are placed on the four limbs of the human body, where the measuring electrodes are at three positions: the right arm (RA), the left arm (LA), and the left leg (LL) of the human body. Different combinations form the signals of leads I, II, and III.

[0148] As Figure 3 shown, this system comprehensively reflects the electrical activity of the heart in the coronal plane and the transverse plane through the electrode layout at different positions on the limbs and the chest.

[0149] The following gives the calculation formulas for each lead of the 12-lead ECG and the Wilson central terminal (WCT):

[0150] Among them, the calculation formulas for bipolar limb leads I, II, and III are:

[0151] Lead I = V LA - V RA (1.2)

[0152] Lead II = V LL - V RA (1.3)

[0153] Lead III = V LL - V LA = Lead II - LeadI (1.4)

[0154] The calculation formulas for augmented limb leads aVR, aVL, and aVF are:

[0155] Lead III = V LL - V LA = Lead II - LeadI (1.5)

[0156] aVL = V LA - (V RA + V LL ) / 2 = LeadI - 0.5 * LeadII (1.6)

[0157] aVF = V LL - (V RA + V LL ) / 2 = LeadII - 0.5 * LeadI (1.7)

[0158] The calculation formula for the Wilson central terminal potential WCT is as follows:

[0159] WCT = (V RA + V LA + V LL ) / 3(1.8)

[0160] The calculation formulas for chest leads V1 - V6 are as follows:

[0161] V i = V cheat,i - WCT, i ∈ {1, 2, 3, 4, 5, 6}(1.9)

[0162] From the above 8 formulas, the electrocardiogram signals of 12 leads can be calculated, including 3 bipolar limb leads, 3 augmented limb leads, and 6 chest leads. Among these 12 electrocardiogram leads, Lead III, aVR, aVL, and aVF can all be calculated from Lead I and Lead II. Therefore, during the actual acquisition process, only the potentials of the electrodes at V LA , V RA , and V LL need to be acquired, and then the electrocardiogram signals of bipolar limb leads and augmented limb leads can be deduced.

[0163] Similarly, for chest leads, after the potentials of the electrodes at V LA , V RA , and V LL have been acquired, the reference potential of WCT can be calculated. Therefore, only by additionally acquiring the potentials at the positions of 6 chest leads, the electrocardiogram signals of the complete 6 chest leads can be obtained. Therefore, among the 12 - lead electrocardiogram signals, a total of 10 electrodes are used, 9 of which are used to record signals, and the electrocardiogram signals of 8 leads (Lead I, Lead II, V1 - V6) are recorded. The electrocardiogram signals of the other 4 leads (Lead III, aVR, aVL, aVF) are obtained through calculation, thus generating 12 electrocardiogram leads, and the RL (right leg) electrode serves as a drive reference electrode, which does not participate in the formation of electrocardiogram signals and is used to reduce common - mode noise and improve signal quality.

[0164] In addition, for the signals of Lead I, Lead II, and the 6 chest leads that need to be measured, there is a more convenient method, which is crucial for the optimization of the circuit. By transforming the relationships between Lead I, Lead II, and V RA , V LA , V LL according to formulas (1.2), (1.3), and (1.8), the WCT can be transformed, and the WCT can be transformed into the following formula:

[0165] WCT = (V RA + Lead I + V RA + Lead II + V RA ) / 3(1.10)

[0166] Finally, WCT can be transformed into:

[0167] WCT = V RA + (Lead I + Lead II) / 3(1.11)

[0168] At this time, from formula (1.9), the calculation formulas for chest leads V1 - V6 can be deduced as follows:

[0169] V i = V cheat,i - V RA - (Lead I + Lead II) / 3, i ∈ {1, 2, 3, 4, 5, 6}(1.12)

[0170] Combining the three formulas (1.2), (1.3), and (1.21), among the 8-lead ECG signals (LeadI, Lead II, V1 - V6) that need to be measured, V RA can be used as the negative pole. Applying this transformation to the front-end acquisition circuit of the ECG can effectively reduce the circuit complexity. Without first constructing the Wilson central terminal potential (WCT) circuit, the negative poles of all differential signals can be directly connected to V RA , and then through simple mathematical calculations in the microcontroller, the ECG signals of each lead in the standard formula can be obtained.

[0171] The filtering unit is a protection circuit that performs energy limitation and low-pass filtering on the input high-frequency ECG signals. The entire circuit design consists of a gas discharge tube (2RK075M - 4), a current-limiting resistor, a filtering capacitor, and a rectifier diode (BAV199). The two ends of the gas discharge tube (2RK075M - 4) are respectively connected to the lead input terminal and the floating ground terminal, and are used to limit the voltage amplitude of the input signal. When the amplitude of the input signal exceeds the set threshold, the gas discharge tube will conduct, transferring the excess energy of the signal to the ground wire, thereby preventing subsequent circuits from being damaged by excessive signals.

[0172] One end of the current-limiting resistor is connected to the lead input terminal, and the other end is connected to the rectifying diode, which is used to limit the current of the input signal and prevent excessive current from affecting the subsequent circuit. One end of the filtering capacitor is connected to the non-inverting terminal of the input buffer circuit, and the other end is grounded to form an RC low-pass filter circuit to perform low-pass filtering on the input signal and achieve high-frequency noise suppression. The low-pass RC filtering parameters set here are resistor R = 44 kΩ and capacitor C = 100 pF. This circuit can effectively attenuate high-frequency signals above 36 kHz. The rectifying diode (BAV199) adopts a center-tapped structure, with the center terminal connected to the non-inverting input terminal of the input buffer circuit, and the other two terminals are respectively connected to the positive and negative power supplies. The main function of this circuit is to rectify the input signal and provide a stable voltage reference for the differential signal processing of the subsequent circuit.

[0173] The buffer unit adopts a unity-gain (Voltage Follower) structure, and its core component is a TLV2221IDBVR operational amplifier with low noise and low offset voltage, aiming to increase the input impedance of the high-frequency electrocardiogram signal, reduce signal attenuation caused by skin-electrode contact impedance, and at the same time provide a low output impedance to ensure stable signal transmission to the subsequent circuit. The circuit is powered by a ±2.5V symmetrical dual power supply to ensure signal integrity, and is equipped with an R25 (18 kΩ) current-limiting resistor and C108, C109 (100 nF) decoupling capacitors to suppress noise and improve power supply stability. The overall design can effectively reduce signal distortion and improve the accuracy of high-frequency electrocardiogram signal acquisition, providing high-quality input for subsequent processing.

[0174] The Right Leg Drive (RLD) circuit is a negative feedback technique used to reduce common-mode noise and improve the signal-to-noise ratio (SNR) of the electrocardiogram signal. Since the electrode signals of the human body are extremely susceptible to power frequency interference (50 Hz / 60 Hz) and other common-mode noise, in order to reduce these interferences, an RLD circuit is usually used to feedback the common-mode signal back to the human body, thereby actively canceling these noises and improving the quality of the electrocardiogram signal.

[0175] The extraction of the common-mode signal in the RLD noise reduction unit generally extracts the common-mode signal from the left and right hand electrodes (RA and LA) or all limb leads (RA, LA, LL), and then the extracted common-mode signal is reversely amplified by an inverting amplifier. After reverse amplification, in order to avoid the introduction of high-frequency noise, an RC low-pass filter is usually added at the output terminal of the RLD circuit to filter out high-frequency noise. The filtered signal is finally fed back to the human body through the right leg electrode (RL), thereby effectively suppressing common-mode interference and improving the accuracy and stability of electrocardiogram signal acquisition.

[0176] The ECG sensor of the present invention is an ADS131E08S chip. With its advantages such as high precision, low noise, and multi-channel synchronous sampling, the ADS131E08S chip can effectively improve the signal quality and provide high-quality input for subsequent data processing, which is very suitable for the requirements of this system.

[0177] ADS131E08S is a highly integrated multi-channel synchronous sampling ADC chip that supports up to 24-bit Δ-Σ (ΔΣ) analog-to-digital conversion and is designed for high-precision signal acquisition. The chip integrates a programmable gain amplifier (PGA), a high-precision reference source, and a low-noise oscillator internally, reducing the dependence on external devices, thereby optimizing the overall performance and power consumption of the system. Its sampling rate can be set in the range of 1 to 64 kSPS through the configuration register, and the PGA gain can be adjusted to 1, 2, 4, 8, or 12 to adapt to input signals of different amplitudes. In addition, the chip has a dynamic range of up to 118 dB at a sampling rate of 1 kSPS, a crosstalk index of -125 dB, a total harmonic distortion (THD) of -100 dB at both 50 Hz and 60 Hz frequencies, and an internal reference voltage drift of only 8 ppm / °C. Its operating temperature range covers -40°C to +105°C, which can meet the environmental requirements of daily ECG signal acquisition.

[0178] After passing through the differential amplification unit, the analog signal is transmitted to the ADS131E08S and amplified by the programmable gain amplifier (PGA) integrated inside the chip. The PGA module supports gain adjustment of 1, 2, 4, 8, and 12 times, and the system noise level is also different under different gain coefficients. Generally speaking, the larger the gain coefficient, the lower the system noise, but the noise level is also affected by the sampling frequency.

[0179] In addition, the ADS131E08S supports power supply of 3V and 5V. Considering the requirements of portable design, this solution selects 3V power supply to reduce power consumption. At this time, the reference voltage is 2.4V, and the amplitude range of the ECG signal is 0.5 to 4 mV. The system noise levels at different sampling frequencies and gains are shown in Table 2.1. In order to reduce noise as much as possible, a higher gain coefficient should be selected. If a 12-fold gain is set, the maximum amplitude of the ECG signal can be amplified to 48 mV, which is much lower than the reference voltage of 2.4V, ensuring that the signal is within a safe range. Therefore, the present invention selects a gain coefficient of 12 to optimize the signal quality and improve the acquisition accuracy.

[0180] After that, the amplified analog signal needs to pass through the ΔΣ analog-to-digital converter (ADC). The ADS31E08S provides seven optional sampling frequencies of 1, 2, 4, 8, 16, 32, and 64 kSPS. Among them, the sampling frequencies of 16 kSPS and below can support a maximum resolution of 24 bits, while 32 kSPS and 64 kSPS only support a maximum resolution of 16 bits.

[0181] Monitoring of high-frequency electrocardiogram (ECG) signals requires capturing more subtle changes in electrical signals. Therefore, the sampling bit number should be increased as much as possible. At a resolution of 24 bits, the maximum supported sampling frequency is 16 kSPS, and the corresponding system - 3 dB cut-off frequency is 4192 Hz. Generally, the frequency range required for high-frequency ECG signal analysis is 150 - 250 Hz. To ensure complete signal acquisition, the - 3 dB cut-off frequency generally needs to reach at least 10 times the maximum analysis frequency. Thus, 4192 Hz ≥ 10×250 Hz, meeting this requirement. Therefore, the present invention adopts a sampling rate of 16 kSPS to achieve the highest sampling efficiency at 24-bit precision, ensuring the integrity and accuracy of high-frequency ECG signals.

[0182] Through acquisition by the 8 differential channels of the AD conversion unit, 8 groups of differential signals can be obtained, where LA, LL, V1 - V6 are the positive electrodes and RA is the negative electrode. According to Formulas (1.2) and (1.3) in Section 2.2, the Lead I and Lead II signals respectively correspond to the digital signals acquired by Channel 1 and Channel 2.

[0183] According to Formula (1.21), the differential signals are acquired by Channels 3 to 8, but there is still a certain deviation between this signal and the standard ECG chest lead signals. Since Lead I and Lead II are respectively acquired by Channel 1 and Channel 2, it is necessary to convert the differential signals at the microcontroller end to obtain the standard chest lead signals.

[0184] In addition, the Lead III, aVR, aVL, and aVF lead signals can be obtained by performing mathematical operations (such as Formulas (1.4) to (1.7)) on Lead I and Lead II.

[0185] The sampling frequency of the ADS131E08S is set to 16 kSPS and supports eight-channel synchronous acquisition. Therefore, the total data transfer volume reaches 16 kSPS×8. The sampling data bit width per channel is 24 bits (3 bytes), and usually a 32-bit format is used for transmission to retain the sign bit and optimize data processing. Since the overall data volume is large, direct transmission will significantly increase the calculation and storage burden of the microcontroller. In addition, although a higher ADC sampling rate can improve the time-domain resolution, it also increases the computational overhead of data processing. Therefore, the original ECG signals can be subjected to average filtering and downsampling to reduce the data volume and improve the signal quality.

[0186] The present invention performs downsampling by averaging 16 groups of data to reduce the data volume, reduce the data transmission burden on the microcontroller, smooth random noise at the same time, and improve the signal-to-noise ratio. After downsampling, the sampling frequency is reduced from 16 kSPS to 1 kSPS. According to the Nyquist sampling theorem, that is, the sampling frequency needs to be at least twice the highest frequency of the signal to avoid aliasing (F_s≥2F_max). The spectral range after downsampling is still 4 times that of the 150 - 250 Hz band of the high-frequency electrocardiogram signal (1 kSPS≥2×250 Hz), which can completely retain the key information of the high-frequency electrocardiogram signal.

[0187] In addition, to filter out the baseline drift interference, the present invention uses a Butterworth digital high-pass filter to filter the data after the conversion of the standard electrocardiogram signal, and deploys the designed digital filter into the MCU. Using the filterDesigner toolbox in Matlab, a 2nd-order IIR high-pass Butterworth filter is designed, with the sampling frequency set to 1 kHz and the cut-off frequency to 0.67 Hz.

[0188] After the data is processed and packed by STM32F405 as the main control chip, the packed data needs to be transmitted. To meet the portable requirements, wireless transmission is required. In the present invention, STM32F072 is used as the wireless communication control unit, and is paired with the HC-04 wireless communication module for wireless data transmission to meet the real-time transmission requirements of the portable high-frequency electrocardiogram signal.

[0189] STM32F072 is a low-power microcontroller based on the ARM Cortex-M0 core, with a main frequency of up to 48 MHz, integrating multiple communication interfaces such as USART, SPI, I 2 C, USB, etc., and having a powerful peripheral control ability. In the system of the present invention, STM32F072 is selected as the wireless transmission control unit, mainly based on the following considerations: First, this chip can independently undertake the wireless transmission task, decouple the data communication from the main control chip STM32F405, reduce the main control calculation burden, and improve the stability of the system; Second, STM32F072 is compatible with a variety of wireless communication modules, supports flexible communication protocols, meets the portable requirements of this system, and optimizes the efficiency and power consumption of wireless data transmission.

[0190] HC-04 is a wireless serial communication module based on the Bluetooth 2.0 protocol. It supports the UART interface and features low power consumption, strong stability, and easy integration. In the system of the present invention, HC-04, as the wireless communication module, is responsible for sending the data processed by STM32F072 to the host computer (such as a PC or a mobile device) via Bluetooth. Its low power consumption characteristic makes it suitable for portable devices, and the simple configuration method of the UART interface facilitates efficient docking with STM32F072 to achieve stable data transmission.

[0191] Among them, HC-04, as the wireless communication module, is mainly responsible for receiving and sending data. The receiving part is used to parse the instructions sent by the host computer, including two instructions: start ("START") and end ("CLOSE"), and transfer the instructions to STM32F072 for it to execute corresponding tasks. In STM32F072, it is mainly responsible for data acquisition and packet integrity judgment, specifically including: (1) Data verification: Verify the checksum and check the data length of the data collected from the main control chip STM32F405 to ensure the integrity and accuracy of the data. (2) Exception handling: If the data is complete, it is sent to the host computer via HC-04; if the data is missing or abnormal, the data is discarded to avoid error propagation.

[0192] The central electrocardiogram sensor in the front-end acquisition and processing module is used to collect high-frequency electrocardiogram signals (150Hz - 250Hz) from the human body surface to capture subtle changes in myocardial activity. The weak high-frequency electrocardiogram signals collected are amplified to ensure the acquisition of signals with high signal-to-noise ratio. The amplified analog signals are converted into digital signals by the sampling rate AD conversion unit to meet the high-frequency characteristic analysis requirements of high-frequency electrocardiogram signals and provide high-quality signal input for subsequent data processing and analysis. The processing process includes operations such as average noise reduction, baseline drift filtering, and low-pass filtering of the electrocardiogram signals to reduce noise interference and ensure that the extracted high-frequency electrocardiogram signals are as pure as possible. In addition, the single-chip microcomputer in the processing module is also responsible for comprehensively controlling the system to ensure the coordinated operation of each module.

[0193] The wireless communication module integrates a Bluetooth module, enabling the device to perform data transmission and real-time communication with mobile devices (such as smartphones, tablets, etc.) or cloud platforms. This function facilitates remote monitoring and storage of data, ensuring that medical staff can always grasp the electrocardiogram status of patients and provide timely diagnosis and intervention.

[0194] The present invention combines the LSTM in the pre-trained electrocardiogram (ECG) signal classification network to capture the time series features of the signal and analyze pathological states such as myocardial ischemia and myocardial infarction in real time. After the training of the ECG signal classification network is completed, through quantization and compression processing, the ECG signal classification network is successfully deployed to a portable device, realizing efficient and low-power dynamic high-frequency ECG monitoring. Compared with traditional methods, the present invention significantly improves the accuracy of high-frequency ECG signal analysis, and its portable feature enables the device to provide early warning of myocardial infarction in daily life. The device does not rely on exercise load as an excitation condition, is more user-friendly for groups with difficulty in moving, helps prevent myocardial infarction, and provides important support for the early diagnosis of cardiovascular diseases.

[0195] The power supply module adopts a high-performance battery module to support long-term portable monitoring. The battery module is optimized to ensure the stability and reliability of the device during long-term use, meet the needs of patients for continuous wearing in daily life, and ensure sufficient battery life for early warning of diseases such as myocardial ischemia and myocardial infarction.

[0196] The present invention uses deep learning technology to train high-frequency ECG signals and deploys the trained model to a portable device. Different from traditional methods, the present invention uses resting high-frequency ECG data for training and does not rely on exercise load at all. When a set number of ECG data is collected, it can be directly input into the trained network to give a diagnosis result in real time, eliminating the lag in the analysis process of traditional methods. Through the automatic feature extraction of the deep learning model, the deficiencies of traditional methods that rely on manually designed feature extraction and classification algorithms are avoided, improving the diagnostic efficiency and accuracy. After being optimized by deep learning, the diagnostic accuracy of electrocardiograms can reach over 90%, significantly improving the early diagnostic ability of myocardial ischemia and myocardial infarction.

[0197] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. All or part of the present invention can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, mobile communication terminals, multi-processor systems, microprocessor-based systems, programmable electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on.

[0198] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting the present invention; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the present invention.

Claims

1. A portable dynamic high-frequency electrocardiogram monitoring method, characterized in that, Including: Obtain resting electrocardiogram data, perform data preprocessing on the resting electrocardiogram data to obtain resting high-frequency electrocardiogram data; Input the resting high-frequency electrocardiogram data into a pre-trained electrocardiogram signal classification network to obtain the monitoring result of the resting high-frequency electrocardiogram for myocardial ischemia; wherein, the electrocardiogram signal classification network includes: a multi-scale convolutional neural module, a feature splicing module, a long short-term memory module, an adaptive convolutional module, and a classification module; The multi-scale convolutional neural module is used to perform multi-scale feature extraction on the resting high-frequency electrocardiogram data to obtain local detail features corresponding to each convolutional layer; The feature splicing module is used to splice the local detail features corresponding to each convolutional layer to obtain an activation fusion splicing feature; The long short-term memory module is used to perform temporal modeling on the activation fusion splicing feature to obtain temporal correlation information; The adaptive convolutional module is used to randomly generate dynamic convolution kernels to perform feature extraction on the temporal correlation information to obtain temporal detail features corresponding to each convolutional layer; The classification module is used to classify the temporal detail features corresponding to each convolutional layer to obtain a classification result.

2. The portable dynamic high-frequency electrocardiogram monitoring method according to claim 1, wherein The step of performing data preprocessing on the resting electrocardiogram data to obtain resting high-frequency electrocardiogram data includes: Split the resting electrocardiogram data to obtain multiple groups of split resting electrocardiogram data; After cleaning the multiple groups of split resting electrocardiogram data, calculate the correlation coefficient between the intra-group data of each group of split resting electrocardiogram data, and determine whether the correlation coefficient is less than a first preset value; If the correlation coefficient between the intra-group data of a split resting electrocardiogram data is less than the first preset value, discard the split resting electrocardiogram data from the multiple groups of split resting electrocardiogram data; For the split resting electrocardiogram data with the correlation coefficient between the intra-group data not less than the first preset value, extract the high-frequency electrocardiogram signal from the split resting electrocardiogram data through a band-pass filter to obtain the resting high-frequency electrocardiogram data.

3. The portable dynamic high-frequency electrocardiogram monitoring method according to claim 1, wherein The multi-scale convolutional neural module includes a first convolutional layer, a second convolutional layer, and a third convolutional layer arranged in parallel; The multi-scale convolutional neural module is used to perform multi-scale feature extraction on the resting high-frequency electrocardiogram data to obtain local detail features corresponding to each convolutional layer, including: According to the first convolutional layer, perform fast fluctuation detail feature extraction on the QRS wave band in the resting high-frequency electrocardiogram data to obtain a first feature; According to the second convolutional layer, perform overall morphology feature extraction on the QRS wave band in the resting high-frequency electrocardiogram data to obtain a second feature; According to the third convolutional layer, perform heart rate transformation and QRS intensity feature extraction on the QRS wave band in the resting high-frequency electrocardiogram data to obtain a third feature; Take the first feature, the second feature, and the third feature as the local detail features corresponding to each convolutional layer.

4. The portable dynamic high-frequency electrocardiogram monitoring method according to claim 3, wherein The feature splicing module includes: a filter splicing layer, a batch normalization layer, and a ReLU activation function layer; the feature splicing module is used to splice the local detail features corresponding to each branch to obtain an activation fusion splicing feature, including: According to the filter splicing layer, splice the first feature, the second feature, and the third feature to obtain an initial activation fusion spliced feature; According to the batch normalization layer, perform a normalization operation on the initial activation fusion spliced feature to obtain a normalized spliced feature; According to the ReLU activation function layer, perform a non-linear operation on the normalized spliced feature to obtain an activation fusion spliced feature.

5. The portable dynamic high-frequency electrocardiogram monitoring method according to claim 1, wherein The long short-term memory module includes a TimeDistributed layer and an LSTM layer connected in sequence; The long short-term memory module is used to perform temporal modeling on the activation fusion spliced feature to obtain temporal correlation information, including: According to the TimeDistributed layer, flatten the activation fusion spliced feature at each time step to obtain a one-dimensional feature vector corresponding to each time step; Use the LSTM layer to perform temporal modeling on the one-dimensional feature vector corresponding to each time step and capture long-term and short-term dependencies to obtain temporal correlation information.

6. The portable dynamic high-frequency electrocardiogram monitoring method according to claim 1, wherein The adaptive convolution module includes a fourth convolution layer, a fifth convolution layer, and a sixth convolution layer arranged in parallel; wherein, the number of convolution kernels of the convolution layers in the fourth convolution layer, the fifth convolution layer, and the sixth convolution layer changes adaptively according to the temporal correlation information; The adaptive convolution module is used to randomly generate dynamic convolution kernels to extract features from the temporal correlation information to obtain temporal detail features corresponding to each convolution layer, including: Extract the small-scale high-frequency transient features of the temporal correlation information according to the fourth convolution layer to obtain the first self-convolution temporal correlation information; Extract the medium-scale intermediate-frequency morphological features of the temporal correlation information according to the fifth convolution layer to obtain the second self-convolution temporal correlation information; Extract the large-scale low-frequency rhythm features of the temporal correlation information according to the sixth convolution layer to obtain the third self-convolution temporal correlation information.

7. The portable dynamic high-frequency electrocardiogram monitoring method according to claim 1, wherein, The classification module includes: a channel feature splicing layer, a batch normalization activation flattening layer, a regularization fully connected activation layer, and a fully connected discrimination layer; The classification module is used to classify the temporal detail features corresponding to each convolution layer to obtain a classification result, including: According to the channel feature splicing layer, splice the temporal detail features corresponding to each convolution layer to obtain an initial spliced self-convolution temporal correlation information; According to the batch normalization activation flattening layer, perform batch normalization, activation, and flattening operations on the initial spliced self-convolution temporal correlation information to obtain a one-dimensional self-convolution temporal feature; According to the regularization fully connected activation layer, perform regularization and compression operations on the one-dimensional self-convolution temporal feature to obtain a compressed self-convolution temporal feature; According to the fully connected discrimination layer, classify the compressed self-convolution temporal feature to obtain a classification result.

8. The portable dynamic high-frequency electrocardiogram monitoring method according to claim 1, wherein The loss function used in the process of training the electrocardiogram signal classification network is: Among them, N represents the total number of training samples; y i represents the true label of the i-th training sample; p i represents the prediction result of the i-th training sample.

9. A portable dynamic high-frequency electrocardiogram monitoring system, characterized in that, For implementing the portable dynamic high-frequency electrocardiogram monitoring method according to any one of claims 1 to 8, the system includes: a front-end acquisition and processing module, a signal primary processing module, a wireless communication module, a result prediction module, and a power supply module; The front-end acquisition and processing module: It is used to collect multi-channel body surface potentials multiple times, and obtain multiple groups of primary high-frequency electrocardiogram signals by performing differential amplification and AD conversion on the high-frequency electrocardiogram signals; The primary signal processing module: It is used to perform summation downsampling on the multiple groups of primary high-frequency electrocardiogram signals to obtain the denoised primary high-frequency electrocardiogram signals; then calculate the denoised primary high-frequency electrocardiogram signals to obtain the standard high-frequency electrocardiogram signals; The wireless communication module: It is used to transmit the standard high-frequency electrocardiogram signals to the corresponding display module according to the wireless transmission function of the integrated Bluetooth module; The result prediction module: It is used to perform result prediction according to the pre-trained electrocardiogram signal classification network and the standard high-frequency electrocardiogram signals to obtain the monitoring results of resting high-frequency electrocardiogram for myocardial ischemia; among them, the pre-trained electrocardiogram signal classification network is deployed on the dynamic high-frequency electrocardiogram monitoring system; The power supply module: It is used to supply power to the front-end acquisition and processing module, the wireless communication module and the primary signal processing module.

10. The portable dynamic high-frequency electrocardiogram monitoring system according to claim 9, characterized in that It includes: The front-end acquisition and processing module includes: an acquisition unit, a filtering unit, a buffering unit, a right leg drive noise reduction unit, a differential amplification unit and an AD conversion unit; The front-end acquisition and processing module: It is used to collect multi-channel body surface potentials multiple times, and obtain multiple groups of primary high-frequency electrocardiogram signals by performing differential amplification and AD conversion on the high-frequency electrocardiogram signals, including: The acquisition unit: It is used to collect multiple groups of multi-channel body surface potentials multiple times by using an electrocardiogram sensor; among them, each group of multi-channel body surface potentials includes: body surface potentials at electrode positions LA, LL, RA and V1-V6; The filtering unit: It is used to perform low-pass filtering on the multi-channel body surface potentials to obtain the low-pass filtered multi-channel body surface potentials; The buffering unit: It is used to provide a low output impedance for the low-pass filtered multi-channel body surface potentials to obtain the buffered multi-channel body surface potentials; The right leg drive noise reduction unit: It is used to perform noise reduction processing on the buffered multi-channel body surface potentials to obtain the noise-reduced multi-channel body surface potentials; The differential amplification unit: It is used to perform differential processing on the noise-reduced multi-channel body surface potentials to obtain multiple groups of differentially amplified signals; The AD conversion unit: It is used to perform AD conversion on the multiple groups of differentially amplified signals to obtain multiple groups of primary high-frequency electrocardiogram signals.

Images