Night non-contact electrocardiosignal classification method
By using a high-precision, lightweight deep learning model to classify non-contact ECG signals at night, the problems of low signal utilization and high computational complexity in nighttime ECG signal monitoring are solved, achieving efficient signal classification and real-time monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGDONG HONG KONG MACAO GREATER BAY AREA PRECISION MEDICINE RESEARCH INSTITUTE (GUANGZHOU)
- Filing Date
- 2024-11-01
- Publication Date
- 2026-05-08
AI Technical Summary
Existing ECG signal quality classification methods are mainly for signals acquired through traditional contact methods. There is a lack of specific design for monitoring ECG signals at night, and non-contact signals are easily affected by environmental noise, resulting in low signal utilization and high model complexity.
A high-precision, lightweight deep learning model is used to collect non-contact ECG signals at night through capacitive coupling. The signals are preprocessed and segmented, and then feature extraction and classification are performed using a convolutional neural network. The signals are divided into four categories: C1, C2, C3, and C4, which are used for arrhythmia diagnosis, heart rate variability calculation, body movement detection, and bed exit recognition.
It improves the utilization rate of ECG signals, reduces computational complexity, and achieves efficient nighttime ECG signal classification, making it suitable for real-time monitoring systems.
Smart Images
Figure CN121997107A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of health monitoring technology, and in particular to a method for classifying non-contact electrocardiogram signals at night. Background Technology
[0002] Currently, there are 330 million people with cardiovascular disease in China, and 10.3% of the population is considered high-risk. The high prevalence of cardiovascular disease leads to a persistently high proportion of deaths. Electrocardiogram (ECG) signal acquisition is one of the important methods for diagnosing cardiovascular disease. Traditional ECG acquisition typically uses Ag / AgCl electrodes. However, long-term ECG monitoring using Ag / AgCl electrodes may irritate the skin and cause dermatitis.
[0003] Therefore, an increasing number of researchers are adopting non-contact methods to collect ECG signals, eliminating the need for direct contact with the subject. However, these methods are susceptible to environmental noise interference, leading to significant signal loss and low signal utilization. Therefore, improving signal utilization through quality classification methods for non-contact ECG signals is crucial. Current ECG signal quality classification methods primarily target signals acquired using traditional methods, often categorizing them as acceptable or unacceptable, resulting in low signal utilization. Furthermore, most methods involve complex models and high computational costs. Finally, nighttime is a high-risk period for acute cardiovascular events, and current ECG signal quality classification methods are not specifically designed for nighttime ECG signal monitoring.
[0004] Therefore, there is an urgent need in this field to develop a non-contact classification method for ECG signals at night. This method improves the utilization rate of ECG signals by using a high-precision, lightweight deep learning model to classify ECG signals, while having low computational complexity and achieving efficient classification. Summary of the Invention
[0005] The purpose of this application is to provide a non-contact classification method for ECG signals at night. This method uses a high-precision, lightweight deep learning model to classify ECG signals, thereby improving the utilization rate of ECG signals. At the same time, it has low computational complexity and achieves efficient classification.
[0006] The first aspect of this application provides a method for training a deep learning classification model for nighttime non-contact electrocardiogram signal classification, comprising the following steps:
[0007] (a) Acquiring non-contact ECG signal data at night based on the principle of capacitive coupling;
[0008] (b) The nighttime non-contact ECG signal data is preprocessed, the preprocessed nighttime non-contact ECG signal data is segmented, and the nighttime non-contact ECG signal data is classified according to the characteristics of each segment, thereby obtaining nighttime non-contact ECG signal data with classification labels.
[0009] (c) Divide the labeled nighttime non-contact ECG signal data into training set, validation set and test set;
[0010] (d) The nighttime non-contact ECG signal data with classification labels in the training set are used as input features of the deep learning classification model, and the classification labels of the nighttime non-contact ECG signal data are used as output features of the deep learning classification model to train the deep learning classification model.
[0011] The deep learning classification model includes a feature extraction module and a classification module composed of a convolutional neural network. The nighttime non-contact ECG signal data with classification labels includes one-dimensional ECG signal data and a corresponding classification label for each segment of nighttime non-contact ECG signal data. The horizontal axis of the one-dimensional ECG signal data represents time, and the vertical axis represents the signal amplitude.
[0012] In another preferred embodiment, the preprocessing in step (b) includes using a bandpass filter to remove baseline offset interference and power line interference from the electrocardiogram signal.
[0013] In another preferred embodiment, in step (b), the preprocessed nighttime non-contact ECG signal data is segmented every 2s-5s to obtain multiple segments of nighttime non-contact ECG signals.
[0014] In another preferred embodiment, the preprocessed nighttime non-contact ECG signal data is segmented every 3 seconds.
[0015] In another preferred embodiment, in step (b), classification labels are added to each segment of nighttime non-contact ECG signal data by manual annotation.
[0016] In another preferred embodiment, the nighttime non-contact ECG signal data in step (a) is acquired via a mattress based on the principle of capacitive coupling.
[0017] In another preferred embodiment, in step (b), the segmented nighttime non-contact ECG signal data is divided into four categories, and the classification labels include C1, C2, C3, and C4. C1 refers to clear ECG signals that can be used for subsequent arrhythmia diagnosis; C2 refers to ECG signals where the P wave and T wave are masked by noise, and only the QRS complex can be clearly identified in the signal, which can be used for subsequent heart rate variability calculation; C3 refers to ECG signals that exhibit irregular large fluctuations and can be used for subsequent body movement detection; and C4 refers to ECG signals that exhibit weak noise or are directly displayed as a horizontal straight line and can be used for subsequent bed exit identification.
[0018] In another preferred embodiment, the feature extraction module includes at least five convolutional blocks, each with the same structure, and each convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer in sequence.
[0019] In another preferred embodiment, the feature extraction module includes a first convolutional block, a second convolutional block, a third convolutional block, a fourth convolutional block, and a fifth convolutional block. Each convolutional block has the same structure, and each convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer in sequence.
[0020] In another preferred embodiment, in step (d), the nighttime non-contact ECG signal data with classification labels is input into the first convolutional block to obtain first ECG signal feature data F1. The first ECG signal feature data F1 is input into the second convolutional block to obtain second ECG signal feature data F2. The second ECG signal feature data F2 is input into the third convolutional block to obtain third ECG signal feature data F3. The third ECG signal feature data F3 is input into the fourth convolutional block to obtain fourth ECG signal feature data F4. The fourth ECG signal feature data F4 is input into the fifth convolutional block to obtain fifth ECG signal feature data F5. The fifth ECG signal feature data F5 is input into the classification module, thereby training the deep learning classification model.
[0021] In another preferred embodiment, the classification module includes a random deactivation layer, a first fully connected layer, an activation function ReLU layer, and a second fully connected layer.
[0022] In another preferred embodiment, the method further includes the following steps:
[0023] (e) Input the one-dimensional data of the electrocardiogram signals in the validation set into the deep learning classification model to obtain the values of the main indicators of the classification performance of the deep learning classification model;
[0024] (f) Then, steps (d)-(e) are executed iteratively to obtain the optimal deep learning classification model;
[0025] (g) Input the one-dimensional data of the electrocardiogram signals in the test set into the optimal deep learning classification model obtained in step (f) to obtain the values of the main indicators of the classification performance of the optimal deep learning classification model.
[0026] In another preferred embodiment, the main performance metrics of the deep learning classification model include recall, precision, F1 score, and accuracy.
[0027] A second aspect of this application provides a method for classifying non-contact electrocardiogram signals at night using the aforementioned deep learning classification model, characterized by comprising the following steps:
[0028] (a) Acquiring non-contact ECG signal data at night based on the principle of capacitive coupling;
[0029] (b) The nighttime non-contact ECG signal data is preprocessed, and the preprocessed nighttime non-contact ECG signal data is segmented to obtain multiple segments of nighttime non-contact ECG signal data.
[0030] (c) Input the nighttime non-contact ECG signal data of each segment into the deep learning classification model to obtain the classification label of the nighttime non-contact ECG signal data of each segment.
[0031] A third aspect of this application provides a system for classifying non-contact electrocardiogram signals at night, including...
[0032] The data acquisition module acquires non-contact ECG signal data from patients at night based on the principle of capacitive coupling.
[0033] The first preprocessing module filters the nighttime non-contact electrocardiogram signal data.
[0034] The second preprocessing module is used to divide the filtered nighttime non-contact ECG signal data every 2s-5s to obtain multiple segments of nighttime non-contact ECG signal data.
[0035] The data feature classification module has a trained deep learning classification model, which is used to classify the multi-segment nighttime non-contact ECG signal data obtained by the second preprocessing module and output classification labels. The trained deep learning classification model is trained by the method of claim 1.
[0036] In another preferred embodiment, the data acquisition module is a textile with a flexible sensor array. Preferably, the data acquisition module includes, but is not limited to, an ECG acquisition mattress, an ECG acquisition sheet, and an ECG acquisition cushion.
[0037] In another preferred embodiment, the first preprocessing module and / or the second preprocessing module are disposed in the acquisition mattress.
[0038] In another preferred embodiment, the first preprocessing module and / or the second preprocessing module are located in a remote processor.
[0039] In another preferred embodiment, the data feature classification module is located in a remote processor.
[0040] In another preferred embodiment, the remote processor is a host computer or a mobile phone.
[0041] A fourth aspect of this application provides a non-contact classification device for nighttime electrocardiogram signals, comprising:
[0042] Memory, used to store computer-executable instructions; and,
[0043] A processor, coupled to the memory, is configured to implement the steps of the above method when executing the computer-executable instructions.
[0044] A fifth aspect of this application provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the method described above.
[0045] A sixth aspect of this application provides a computer program product including computer-executable instructions, characterized in that the computer-executable instructions, when executed by a processor, implement the steps in the above-described method.
[0046] It should be understood that, within the scope of this invention, the above-described technical features of this invention and the technical features specifically described below (such as in the embodiments) can be combined with each other to form new or preferred technical solutions. Due to space limitations, they will not be described in detail here. Attached Figure Description
[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. It should be understood that the accompanying drawings described below are merely some implementation examples of the present invention, and those skilled in the art can obtain other implementation examples based on these drawings without creative effort.
[0048] Figure 1This is a flowchart of a method for classifying non-contact electrocardiogram signals at night according to one embodiment of this application.
[0049] Figure 2 This is a flowchart of a deep learning method according to one embodiment of this application.
[0050] Figure 3 These are schematic diagrams of signal waveforms of Class C1, Class C2, Class C3, and Class C4, where (a) is a Class C1 signal waveform, (b) is a Class C2 signal waveform, (c) is a Class C3 signal waveform, and (d) is a Class C4 signal waveform.
[0051] Figure 4 This is a graph showing the relationship between the number of iterations and the accuracy of the deep learning classification model according to the embodiments of this application;
[0052] Figure 5 This is a schematic diagram of a non-contact ECG signal classification system for nighttime acquisition according to one embodiment of this application. Detailed Implementation
[0053] Through extensive and in-depth research, the inventors have developed for the first time a classification method for non-contact ECG signals collected at night. This method categorizes non-contact ECG signals acquired via a mattress into four types, which are then used for subsequent arrhythmia diagnosis, heart rate variability calculation, body movement detection, and bed exit identification. This method improves signal utilization. Furthermore, the classification method of this application employs a lightweight model with low computational complexity, making it applicable to real-time monitoring systems.
[0054] In the following description, many technical details are presented to help the reader better understand this application. However, those skilled in the art will understand that the technical solutions claimed in this application can be implemented even without these technical details and various variations and modifications based on the following embodiments.
[0055] the term
[0056] It should be noted that in this patent application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. In this patent application, if it refers to performing an action according to an element, it means performing the action at least according to that element, including two cases: performing the action only according to that element, and performing the action according to that element and other elements. Expressions such as "multiple," "repeatedly," and "various" include two, two times, two kinds, and more than two, more than two times, and more than two kinds.
[0057] In this application, all directional indicators (such as up, down, left, right, front, back, etc.) are only used to explain the relative positional relationship and movement of the components in a specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.
[0058] This application possesses at least one of the following advantages.
[0059] (a) The classification method for non-contact ECG signals at night in this application classifies non-contact ECG signals at night into four categories, thereby improving the utilization rate of ECG signals.
[0060] (b) The classification method for non-contact ECG signals at night in this application adopts a lightweight model with low computational complexity and can be applied to real-time monitoring systems.
[0061] (c) The test results obtained by the model used in the classification method of non-contact ECG signals at night in this application are good and the performance index is high, indicating that the model has high accuracy, that is, it provides a high-precision model.
[0062] A training method for a deep learning classification model for non-contact ECG signal classification at night.
[0063] This application provides a training method for a deep learning classification model for non-contact ECG signal quality classification at night, the specific steps of which are as follows:
[0064] Step 1: Acquire raw electrocardiogram (ECG) signals using non-contact capacitive coupling ECG acquisition technology;
[0065] Step 2: Use a bandpass filter to remove baseline offset interference and power line interference from the ECG signal;
[0066] Step 3: Segment the filtered ECG signal every 2-5 seconds and assign a corresponding classification label to each segment based on its characteristics; each segment has only one label.
[0067] Step 4: Divide the segmented signal into training set, validation set and test set, for example, in the form of 8:1:1 and 6:2:2.
[0068] Step 5: Establish a feature extraction module composed of a convolutional neural network, and input the training set signal into the feature extraction module to obtain signal features.
[0069] Step 6: Input the generated signal features into the classification module to obtain the classification model. See [link / reference] Figure 2 .
[0070] Step 7: Input the validation set into the classification model to obtain the values of the main performance indicators of the classification model for the validation set.
[0071] Step 8: Iterate through steps 5-7 multiple times to obtain the best classification model, i.e., the model with the highest accuracy on the validation set. Figure 4 It provides a graph showing the relationship between the number of iterations and the accuracy of the deep learning classification model. Figure 4 The results show that 100 iterations can achieve good results.
[0072] Step 9: Feed the test set into the best classification model to obtain the values of the main indicators of the final classification performance of the classification model.
[0073] In this application, in step 1, the raw ECG signals collected by the non-contact ECG acquisition mattress totaled 48 hours, with a sampling rate of 500Hz. The 48 hours refers to monitoring the same patient for 8 hours each night for 6 nights.
[0074] In this application, in step 2, the bandpass filter used has a cutoff frequency of 0.5Hz and 35Hz, which can remove power line interference while removing baseline interference.
[0075] In this application, in step 3, the signal classification labels include four types: C1, C2, C3, and C4. Among them, C1 is an ECG signal with no visible noise or low noise that does not affect the identification of ECG waveform features. That is, all ECG waveform features, including P waves, QRS complexes, and T waves, can be clearly identified in the signal and are used for subsequent arrhythmia diagnosis. C2 is an ECG signal with a high level of noise, in which the P waves and T waves are masked by noise, and only the QRS complexes can be clearly identified in the signal and are used for subsequent heart rate variability calculation. C3 signals are characterized by irregular large fluctuations and are used for subsequent body movement detection. C4 signals are characterized by weak noise or are directly displayed as a flat line and are used for subsequent bed exit identification.
[0076] In this application, in step 5, the feature extraction module consists of a first convolutional block, a second convolutional block, a third convolutional block, a fourth convolutional block, and a fifth convolutional block.
[0077] The first convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer. The filtered and segmented electrocardiogram signal is input into the first convolutional block to obtain the corresponding features.
[0078] The second convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max-pooling downsampling layer. The features obtained from the first convolutional block are input into the second convolutional block to obtain the corresponding features.
[0079] The third convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max-pooling downsampling layer. The features obtained from the second convolutional block are input into the third convolutional block to obtain the corresponding features.
[0080] The fourth convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer. The features obtained from the third convolutional block are input into the fourth convolutional block to obtain the corresponding features.
[0081] The fifth convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer. The features obtained from the fourth convolutional block are input into the fifth convolutional block to obtain the corresponding features.
[0082] In the first convolutional block, the number of input channels of the one-dimensional convolutional layer is 1, the number of output channels is 16-32, the kernel size is 5-11, the stride is 1-4, and the padding is 1-2; the kernel size of the one-dimensional max pooling downsampling layer in the first convolutional block is 2, and the stride is 2.
[0083] The one-dimensional convolutional layer in the second convolutional block has 16-32 input channels, 16-32 output channels, a kernel size of 3-5, a stride of 1-2, and padding of 1-2; the one-dimensional max-pooling downsampling layer in the second convolutional block has a kernel size of 2 and a stride of 2.
[0084] The one-dimensional convolutional layer in the third convolutional block has 16-32 input channels, 32-64 output channels, a kernel size of 3-5, a stride of 1-2, and padding of 1-2; the one-dimensional max-pooling downsampling layer in the third convolutional block has a kernel size of 2 and a stride of 2.
[0085] The first-dimensional convolutional layer in the fourth convolutional block has 32-64 input channels, 32-64 output channels, a kernel size of 3-5, a stride of 1-2, and padding of 1-2. The first-dimensional max-pooling downsampling layer in the fourth convolutional block has a kernel size of 2 and a stride of 2.
[0086] The five convolutional blocks contain one-dimensional convolutional layers with 32-64 input channels, 32-64 output channels, kernel sizes of 3-5, strides of 1-2, and padding of 1-2. The five convolutional blocks also contain one-dimensional max-pooling downsampling layers with kernel sizes of 2 and strides of 2.
[0087] In this application, in step 6, the classification module consists of a random deactivation layer, a fully connected layer 1, an activation function ReLU layer, and a fully connected layer 2.
[0088] The p-value of the random deactivation layer is set to 0.3-0.5, the number of output channels of fully connected layer 1 is 32-64, and the number of output channels of fully connected layer 2 is 4.
[0089] In this application, in step 9, recall, precision, F1 score, and accuracy are used to evaluate the classification results of the test set to measure the effectiveness of the model.
[0090] A method for nighttime non-contact electrocardiogram signal classification using the aforementioned deep learning classification model.
[0091] This application provides a method for classifying nighttime non-contact electrocardiogram signals using the aforementioned trained deep learning classification model, including the following steps:
[0092] (a) Acquiring non-contact ECG signal data at night based on the principle of capacitive coupling;
[0093] (b) The nighttime non-contact ECG signal data is preprocessed, and the preprocessed nighttime non-contact ECG signal data is segmented to obtain multiple segments of nighttime non-contact ECG signal data.
[0094] (c) Input the nighttime non-contact ECG signal data of each segment into the deep learning classification model to obtain the classification label of the nighttime non-contact ECG signal data of each segment.
[0095] A system for classifying non-contact ECG signals at night
[0096] See Figure 5 A schematic diagram of a non-contact ECG signal classification system for nighttime acquisition is provided. The system includes an ECG acquisition mattress located on a bed, which is wired to an ECG acquisition circuit board. The mattress acquires nighttime non-contact ECG signal data from a patient lying on it based on capacitive coupling. The nighttime non-contact ECG signal data is wirelessly transmitted via Wi-Fi to a host computer or mobile phone. The host computer or mobile phone displays the ECG signals in real time and inputs the signals into a trained deep learning model to output the current ECG signal quality classification result in real time.
[0097] Preferably, the system may further include
[0098] The first preprocessing module filters the nighttime non-contact electrocardiogram signal data.
[0099] The second preprocessing module is used to divide the filtered nighttime non-contact ECG signal data every 2s-5s to obtain multiple segments of nighttime non-contact ECG signal data.
[0100] The data feature classification module has a trained deep learning classification model, which is used to classify the multi-segment nighttime non-contact ECG signal data obtained by the second preprocessing module and output classification labels. The trained deep learning classification model is trained by the method of claim 1.
[0101] Preferably, the first preprocessing module and / or the second preprocessing module are disposed in the acquisition mattress.
[0102] Preferably, the first preprocessing module and / or the second preprocessing module are located in a remote processor.
[0103] Preferably, the data feature classification module is located in a remote processor.
[0104] Preferably, the remote processor is a host computer or a mobile phone.
[0105] To make the objectives, technical solutions, and advantages of the present invention clearer, embodiments of the present invention will be described in further detail below with reference to the accompanying drawings. It should be understood that these are merely examples provided to the reader of possible implementations of the present invention and are not intended to limit the scope of the invention.
[0106] Example 1
[0107] The ECG signal quality classification method of this application first filters and preprocesses the ECG waveform collected from the mattress throughout the night, then divides the signal into 3-second segments, and finally feeds the segmented ECG signal into a lightweight deep learning network for quality classification, and outputs four classification results for subsequent arrhythmia diagnosis, heart rate variability calculation, body movement detection, and bed exit recognition.
[0108] The flowchart of this application is as follows: Figure 1 As shown, it includes the following steps:
[0109] Step 1: Collect raw ECG signals throughout the night using a non-contact ECG acquisition mattress;
[0110] Step 2: Use a bandpass filter with a passband range of 0.5-35Hz to remove baseline offset interference and electric field interference from the ECG signal, and perform filtering preprocessing on the original ECG signal;
[0111] Step 3: The filtered ECG signal is segmented every 3 seconds, resulting in 56,230 signals. Each segment is then labeled with a corresponding classification tag based on its signal characteristics, categorized into four classes: C1, C2, C3, and C4. Class C1 signals are those with no visible noise or low noise that does not affect the identification of ECG waveform features; all waveform features, including the P wave, QRS complex, and T wave, are clearly identifiable in the signal, and are used for subsequent arrhythmia diagnosis. Class C2 signals have high levels of noise, where the P wave and T wave are masked by noise, and only the QRS complex is clearly identifiable in the signal, and are used for subsequent heart rate variability calculation. Class C3 signals exhibit irregular large fluctuations and are used for subsequent body movement detection. Class C4 signals exhibit weak noise or appear as a flat line and are used for subsequent bed exit identification.
[0112] The waveform diagrams for these four types of signals are as follows: Figure 3 As shown, (a) shows the waveform of a C1 signal, (b) shows the waveform of a C2 signal, (c) shows the waveform of a C3 signal, and (d) shows the waveform of a C4 signal.
[0113] Step 4: Divide the segmented signals into training, validation, and test sets in an 8:1:1 ratio.
[0114] Step 5: Establish a feature extraction module composed of a convolutional neural network. The training set signal is fed into the feature extraction module to obtain signal features. The feature extraction module consists of a first convolutional block, a second convolutional block, a third convolutional block, a fourth convolutional block, and a fifth convolutional block.
[0115] The first convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer. The filtered and segmented electrocardiogram signal is input into the first convolutional block to obtain the corresponding features.
[0116] The second convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max-pooling downsampling layer. The features obtained from the first convolutional block are input into the second convolutional block to obtain the corresponding features.
[0117] The third convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max-pooling downsampling layer. The features obtained from the second convolutional block are input into the third convolutional block to obtain the corresponding features.
[0118] The fourth convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer. The features obtained from the third convolutional block are input into the fourth convolutional block to obtain the corresponding features.
[0119] The fifth convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer. The features obtained from the fourth convolutional block are input into the fifth convolutional block to obtain the corresponding features.
[0120] In the first convolutional block, the one-dimensional convolutional layer has 1 input channel, 16 output channels, a kernel size of 11, a stride of 4, and padding of 2; the one-dimensional max pooling downsampling layer in the first convolutional block has a kernel size of 2 and a stride of 2.
[0121] The one-dimensional convolutional layer in the second convolutional block has 16 input channels, 32 output channels, a kernel size of 5, a stride of 1, and padding of 2; the one-dimensional max-pooling downsampling layer in the second convolutional block has a kernel size of 2 and a stride of 2.
[0122] The one-dimensional convolutional layer in the third convolutional block has 32 input channels, 64 output channels, a kernel size of 3, a stride of 1, and padding of 1; the one-dimensional max-pooling downsampling layer in the third convolutional block has a kernel size of 2 and a stride of 2.
[0123] The one-dimensional convolutional layer in the fourth convolutional block has 64 input channels, 64 output channels, a kernel size of 3, a stride of 1, and padding of 1; the one-dimensional max-pooling downsampling layer in the fourth convolutional block has a kernel size of 2 and a stride of 2.
[0124] The one-dimensional convolutional layer in the fifth convolutional block has 64 input channels, 64 output channels, a kernel size of 3, a stride of 1, and padding of 1; the one-dimensional max pooling downsampling layer in the fifth convolutional block has a kernel size of 2 and a stride of 2.
[0125] Step 6: Input the generated signal features into the classification module to obtain the classification model. The classification module consists of a random deactivation layer, a fully connected layer 1, a ReLU activation function layer, and a fully connected layer 2.
[0126] The p-value of the random deactivation layer is set to 0.5, the number of output channels of fully connected layer 1 is 64, and the number of output channels of fully connected layer 2 is 4.
[0127] Step 7: Feed the validation set into the classification model to obtain the classification result of the validation set.
[0128] Step 8: Iterate through steps 5-7 multiple times to obtain the best classification model, i.e., the model with the highest accuracy on the validation set. Figure 4 The graph shows the relationship between the number of iterations and the accuracy of the deep learning classification model. In this example, 100 iterations were performed. Figure 4 The results show that 100 iterations can achieve good results.
[0129] Step 9: Feed the test set into the best classification model and evaluate the classification results of the test set using recall, precision, F1 score, and accuracy to measure the model's performance. Specific results are shown in Table 1.
[0130] Table 1 Model Performance Indicators
[0131]
[0132] The results show that the ECG signal quality classification method proposed in this application has achieved good results in all dimensions.
[0133] Application of four types of electrocardiogram signals
[0134] The classification labels for nocturnal non-contact ECG signals include four types: C1, C2, C3, and C4. Nocturnal non-contact ECG signals are basically divided into the four categories (C1-C4) mentioned in this application.
[0135] Class C1 is an ECG signal with no visible noise or low noise that does not affect the identification of ECG waveform features. That is, all ECG waveform features, including P wave, QRS complex and T wave, can be clearly identified in the signal and used for subsequent arrhythmia diagnosis.
[0136] Cardiac arrhythmia refers to a disruption of the heart's normal electrical activity, leading to an abnormal heart rhythm. It can manifest as a rapid heartbeat, a slow heartbeat, or an irregular heartbeat, and mainly includes types such as premature atrial contractions (PACs), premature ventricular contractions (PVCs), atrial fibrillation, sinus bradycardia, and sinus tachycardia.
[0137] Class C2 is an ECG signal with a high level of noise, in which the P wave and T wave are masked by noise, and only the QRS complex can be clearly identified in the signal for subsequent heart rate variability calculation.
[0138] Heart rate variability refers to the fluctuation in the time interval between each heartbeat (i.e., the interval between heartbeats). It reflects the interaction between the sympathetic nervous system (which promotes a faster heart rate) and the parasympathetic nervous system (which promotes a slower heart rate). It includes multiple indicators, such as the standard deviation of the interval (SDNN) and the root mean square of the difference between adjacent intervals (RMSSD), etc.
[0139] Class C3 signals exhibit irregular, large fluctuations and are used for subsequent body motion detection;
[0140] Among them, the C3 signal is caused by the relative displacement between the human body and the ECG acquisition mattress, resulting in irregular and huge fluctuations in the acquired signal. Therefore, body movement can be detected through signal characteristics.
[0141] Class C4 signals appear as weak noise or as a flat line, and are used for subsequent bed-off identification.
[0142] Among them, C4 signals are signals collected when the human body does not come into contact with the ECG acquisition mattress, resulting in weak noise or a flat line. Therefore, the signal characteristics can be used to identify when someone leaves the bed.
[0143] Comparative Example 1
[0144] In Comparative Example 1, the applicant removed the fourth and fifth convolutional blocks from Example 1, while keeping the rest of the model unchanged. The same steps as in Example 1 were then used to train and test the model, and the final test results are shown in Table 2. It can be seen that changing the number of convolutional blocks in the feature extraction module of the model reduces the model's accuracy.
[0145] Table 2 Model Performance Indicators
[0146]
[0147] By comparing the results of Example 1 and Comparative Example 1, it is shown that the performance of the model decreased to some extent after changing the number of convolutional blocks.
[0148] The specification of this application contains numerous technical features distributed across various technical solutions. Listing all possible combinations of these technical features (i.e., technical solutions) would make the specification excessively lengthy. To avoid this problem, the various technical features disclosed in the above-described invention, the various technical features disclosed in the following embodiments and examples, and the various technical features disclosed in the accompanying drawings can be freely combined to form various new technical solutions (all of which are considered to have been described in this specification), unless such a combination of technical features is technically infeasible. For example, one example discloses feature A+B+C, and another example discloses feature A+B+D+E. Features C and D are equivalent technical means that serve the same function, and technically only one needs to be used; they cannot be used simultaneously. Feature E can technically be combined with feature C. Therefore, the solution A+B+C+D should not be considered as described because it is technically infeasible, while the solution A+B+C+E should be considered as described.
[0149] All documents mentioned in this application are considered to be incorporated in their entirety into the disclosure of this application so that they can serve as a basis for modifications if necessary. Furthermore, it should be understood that after reading the foregoing disclosure of this application, those skilled in the art can make various alterations or modifications to this application, and these equivalent forms also fall within the scope of protection claimed in this application.
Claims
1. A training method for a deep learning classification model for non-contact ECG signal classification at night, characterized in that, Includes the following steps: (a) Acquiring non-contact ECG signal data at night based on the principle of capacitive coupling; (b) The nighttime non-contact ECG signal data is preprocessed, the preprocessed nighttime non-contact ECG signal data is segmented, and the nighttime non-contact ECG signal data is classified according to the characteristics of each segment, thereby obtaining nighttime non-contact ECG signal data with classification labels. (c) Divide the labeled nighttime non-contact ECG signal data into training set, validation set and test set; (d) The nighttime non-contact ECG signal data with classification labels in the training set are used as input features of the deep learning classification model, and the classification labels of the nighttime non-contact ECG signal data are used as output features of the deep learning classification model. The deep learning classification model is trained to obtain the trained deep learning classification model. The deep learning classification model includes a feature extraction module and a classification module composed of a convolutional neural network. The nighttime non-contact ECG signal data with classification labels includes one-dimensional ECG signal data and a corresponding classification label for each segment of nighttime non-contact ECG signal data. The horizontal axis of the one-dimensional ECG signal data represents time, and the vertical axis represents the signal amplitude.
2. The method as described in claim 1, characterized in that, In step (b), the segmented nighttime non-contact ECG signal data is divided into four categories. The classification labels include C1, C2, C3, and C4. C1 refers to clear ECG signals that can be used for subsequent arrhythmia diagnosis. C2 refers to ECG signals where the P wave and T wave are masked by noise, and only the QRS complex can be clearly identified in the signal, which can be used for subsequent heart rate variability calculation. C3 refers to ECG signals that exhibit irregular large fluctuations and can be used for subsequent body movement detection. C4 refers to ECG signals that exhibit weak noise or are directly displayed as a horizontal straight line and can be used for subsequent bed exit identification.
3. The method as described in claim 1, characterized in that, The feature extraction module includes at least five convolutional blocks, each with the same structure. Each convolutional block consists of a one-dimensional convolutional layer, a ReLU activation function layer, and a one-dimensional max pooling downsampling layer.
4. The method as described in claim 1, characterized in that, The classification module includes a random deactivation layer, a first fully connected layer, a ReLU activation function layer, and a second fully connected layer.
5. The method as described in claim 1, characterized in that, It also includes the following steps: (e) Input the one-dimensional data of the electrocardiogram signals in the validation set into the deep learning classification model to obtain the values of the main indicators of the classification performance of the deep learning classification model; (f) Then, steps (d)-(e) are executed iteratively to obtain the optimal deep learning classification model; (g) Input the one-dimensional data of the electrocardiogram signals in the test set into the optimal deep learning classification model obtained in step (f) to obtain the values of the main indicators of the classification performance of the optimal deep learning classification model.
6. A method for classifying non-contact electrocardiogram signals at night using the deep learning classification model of any one of claims 1-5, characterized in that, Includes the following steps: (a) Acquiring non-contact ECG signal data at night based on the principle of capacitive coupling; (b) The nighttime non-contact ECG signal data is preprocessed, and the preprocessed nighttime non-contact ECG signal data is segmented to obtain multiple segments of nighttime non-contact ECG signal data. (c) Input the nighttime non-contact ECG signal data of each segment into the deep learning classification model to obtain the classification label of the nighttime non-contact ECG signal data of each segment.
7. A system for classifying non-contact electrocardiogram (ECG) signals at night, characterized in that, include The data acquisition module acquires non-contact ECG signal data from patients at night based on the principle of capacitive coupling. The first preprocessing module filters the nighttime non-contact electrocardiogram signal data. The second preprocessing module is used to divide the filtered nighttime non-contact ECG signal data every 2s-5s to obtain multiple segments of nighttime non-contact ECG signal data. The data feature classification module has a trained deep learning classification model, which is used to classify the multi-segment nighttime non-contact ECG signal data obtained by the second preprocessing module and output classification labels. The trained deep learning classification model is trained by the method of claim 1.
8. A non-contact electrocardiogram (ECG) signal classification device for nighttime, characterized in that, include: Memory is used to store executable instructions for a computer; as well as, A processor, coupled to the memory, is configured to implement the steps of the method as described in claim 6 when executing the computer-executable instructions.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions that, when executed by a processor, implement the steps of the method as described in claim 6.
10. A computer program product comprising computer-executable instructions, characterized in that, When the computer-executable instructions are executed by the processor, they implement the steps of the method of claim 6.