Channel attention based anti-photoelectric isolation noise electrocardio monitoring system and method
By using a channel attention-based anti-optical isolation noise ECG monitoring system, the problem of noise interference in optical isolation acquisition is solved, achieving efficient and robust ECG signal diagnosis and reducing the misdiagnosis rate.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF SCI & TECH
- Filing Date
- 2026-02-05
- Publication Date
- 2026-06-09
AI Technical Summary
Existing electrocardiogram (ECG) monitoring systems suffer from severe noise interference during photoelectric isolation acquisition, resulting in a low signal-to-noise ratio, difficulty in accurately extracting weak pathological features, and consequently, a high misdiagnosis rate and poor generalization ability.
An anti-optical-isolation noise ECG monitoring system based on channel attention is adopted, including an optical-isolation acquisition module, a data preprocessing module, a data augmentation module, and an intelligent processing module. Through adaptive weight allocation, fractional Fourier transform, and one-dimensional convolutional neural network, noise-related feature channels are suppressed and pathology-related feature channels are enhanced.
It improves the robustness and diagnostic accuracy of electrocardiogram signals, reduces the misdiagnosis rate, and enhances the efficiency of cardiovascular disease screening and diagnosis.
Smart Images
Figure CN122163224A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of intelligent medical and signal processing technology, and in particular to a channel attention-based electrocardiogram monitoring system and method for resisting photoelectric isolation noise. Background Technology
[0002] Electrocardiogram (ECG) monitoring is a non-invasive diagnostic method that continuously monitors surface ECG signals to assess cardiac function. It is an important tool for diagnosing cardiovascular diseases, based on the principle of detecting changes in the potential difference between two specific points in the body. ECG acquisition devices using opto-isolation technology can effectively ensure the electrical safety of patients and suppress common-mode interference. However, optocouplers themselves can introduce specific noises such as shot noise, and the complex electromagnetic interference in the clinical environment leads to low signal-to-noise ratios and baseline drift in the acquired signals. More importantly, certain pathological features are easily confused in a noisy background. For example, the broad waveform of left bundle branch block can often mask subtle ST segment changes in acute myocardial infarction, resulting in high misdiagnosis rates and poor generalization ability for traditional algorithms and deep learning models that do not consider hardware noise characteristics.
[0003] Therefore, there is an urgent need to invent an electrocardiogram monitoring scheme that can robustly extract features from the noise characteristics of opto-isolated acquisition links, so as to effectively suppress opto-isolated acquisition noise interference, accurately extract weak pathological features, improve the efficiency of cardiovascular disease screening and diagnosis, and reduce the misdiagnosis rate. Summary of the Invention
[0004] The purpose of this invention is to provide an electrocardiogram monitoring system and method that can effectively suppress noise interference from photoelectric isolation acquisition and accurately extract weak pathological features.
[0005] The technical solution to achieve the purpose of this invention is: a channel attention-based anti-photoelectric isolation noise ECG monitoring system, comprising a photoelectric isolation acquisition module, a data preprocessing module, a data enhancement module, an intelligent processing module, and a diagnostic output module;
[0006] The opto-isolated acquisition module is used to acquire multi-lead electrocardiogram signals;
[0007] The data preprocessing module is used to perform adaptive length alignment and standardization processing on the acquired signals;
[0008] The data enhancement module uses a photoelectric shot noise model based on Poisson distribution to perform data enhancement processing on the acquired signal;
[0009] The intelligent processing module assigns adaptive weights to different feature channels of the signal, suppresses noise-related feature channels, and enhances pathology-related feature channels.
[0010] The diagnostic output module is used to output diagnostic results and visual evidence.
[0011] Furthermore, the intelligent processing module includes a fractional Fourier transform layer and a one-dimensional convolutional neural network.
[0012] Furthermore, the fractional Fourier transform layer is placed at the front end of the one-dimensional convolutional neural network to better focus on non-stationary features.
[0013] Furthermore, the one-dimensional convolutional neural network integrates a channel attention module, which employs a Squeeze-and-Excitation module.
[0014] Furthermore, the one-dimensional convolutional neural network employs a one-dimensional residual network.
[0015] Furthermore, the one-dimensional convolutional neural network is trained using a weighted cross-entropy loss function, and the weight of each category of samples in the training set is inversely proportional to the number of samples in that category.
[0016] A channel attention-based method for electrocardiogram monitoring against photoelectric isolation noise includes the following steps:
[0017] Step 1: Construct an intelligent processing module, including a fractional Fourier transform layer and a one-dimensional convolutional neural network, and train the one-dimensional convolutional neural network using a dataset.
[0018] Step 2: Acquire multi-lead ECG signals using an opto-isolated acquisition module;
[0019] Step 3: Perform adaptive length alignment and standardization preprocessing on the acquired signals;
[0020] Step 4: Construct a photoelectric shot noise model and perform data augmentation on the acquired signals;
[0021] Step 5: Input the enhanced signal into the intelligent processing module for feature extraction;
[0022] Step 6: Based on the extracted features, output the ECG diagnosis results.
[0023] Furthermore, the intelligent processing module described in step 1 includes a fractional Fourier transform layer and a one-dimensional convolutional neural network. The one-dimensional convolutional neural network is trained using a dataset, as detailed below:
[0024] Step 1.1: Construct an intelligent processing module, including a fractional Fourier transform layer and a one-dimensional convolutional neural network;
[0025] Step 1.2: Construct the dataset and divide it into training, validation, and test sets in a 7:2:1 ratio;
[0026] Step 1.3: Train the one-dimensional convolutional neural network using the weighted cross-entropy loss function and the Adam optimizer until the accuracy on the validation set converges.
[0027] Furthermore, step 4 involves constructing a photoelectric shot noise model to enhance the acquired signal, as detailed below:
[0028] Step 4.1: Construct a photoelectric shot noise model. Treat the normalized ECG signal as simulated light intensity. Based on the Poisson distribution characteristics of photon counting, generate random noise whose variance is proportional to the instantaneous intensity of the signal. Superimpose it onto the original signal. By adjusting the noise intensity coefficient, simulate the acquisition environment with different signal-to-noise ratios.
[0029] Step 4.2: In each training batch, the collected signals are augmented with a set probability to enable the model to learn the essential structural features of the signals.
[0030] A computer device includes a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the aforementioned channel attention-based anti-optical-isolation noise ECG monitoring method.
[0031] Compared with the prior art, the significant advantages of this invention are:
[0032] (1) This invention introduces the channel attention mechanism into the processing of photoelectric isolated ECG signals, enabling the neural network to simulate an "adaptive filter" that automatically evaluates and suppresses channels heavily contaminated by noise at the feature channel level, while enhancing channels carrying key pathological information. It assigns adaptive weights to different feature channels of the signal, suppresses noise-related feature channels, and enhances pathology-related feature channels, thereby improving the robustness of the features from the source.
[0033] (2) A photoelectric noise model based on Poisson distribution was introduced for data augmentation during training. A learnable fractional Fourier transform layer was introduced at the front end of the network to better focus on non-stationary features, which enhanced the network’s noise resistance and improved the accuracy and robustness of ECG diagnosis in noisy environments.
[0034] (3) It can be directly applied to the intelligent upgrade of existing photoelectric electrocardiogram monitoring equipment or integrated into the new generation of electrocardiogram diagnostic equipment, which has significant practical value for improving the efficiency of cardiovascular disease screening and diagnosis and reducing the misdiagnosis rate. Attached Figure Description
[0035] Figure 1 This is a schematic diagram of the structure of an electrocardiogram monitoring system based on channel attention that is resistant to photoelectric isolation noise according to the present invention.
[0036] Figure 2 This is a schematic diagram of the structure of a one-dimensional residual network integrating the SE attention module in an embodiment of the present invention.
[0037] Figure 3 This is a diagram showing the experimental results of the confusion matrix in an embodiment of the present invention.
[0038] Figure 4 This is a demonstration of the diagnostic system software integrating the above-mentioned algorithms deployed on a computing platform in an embodiment of the present invention.
[0039] Figure 5 This is a graph showing the evaluation results performed on an independent test set in an embodiment of the present invention. Detailed Implementation
[0040] It is readily understood that, based on the technical solution of this invention, those skilled in the art can conceive of various embodiments of this invention without altering its essential spirit. Therefore, the following specific embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of this invention or as limitations or restrictions on its technical solution.
[0041] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the invention.
[0042] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.
[0043] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.
[0044] In all the examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.
[0045] Combination Figure 1 The present invention discloses an electrocardiogram monitoring system based on channel attention that is resistant to photoelectric isolation noise, comprising a photoelectric isolation acquisition module, a data preprocessing module, a data enhancement module, an intelligent processing module, and a diagnostic output module;
[0046] The opto-isolated acquisition module is used to acquire multi-lead electrocardiogram signals;
[0047] The data preprocessing module is used to perform adaptive length alignment and standardization processing on the acquired signals;
[0048] The data enhancement module uses a photoelectric shot noise model based on Poisson distribution to perform data enhancement processing on the acquired signal;
[0049] The intelligent processing module assigns adaptive weights to different feature channels of the signal, suppresses noise-related feature channels, and enhances pathology-related feature channels.
[0050] The diagnostic output module is used to output diagnostic results and visual evidence.
[0051] As a specific example, the intelligent processing module includes a fractional Fourier transform layer and a one-dimensional convolutional neural network.
[0052] As a specific example, the fractional Fourier transform layer is placed at the front end of a one-dimensional convolutional neural network to better focus on non-stationary features.
[0053] As a specific example, the one-dimensional convolutional neural network integrates a channel attention module, which employs a Squeeze-and-Excitation module.
[0054] As a specific example, the one-dimensional convolutional neural network employs a one-dimensional residual network.
[0055] As a specific example, the one-dimensional convolutional neural network is trained using a weighted cross-entropy loss function, where the weight of each category of samples in the training set is inversely proportional to the number of samples in that category.
[0056] This invention also provides a method for electrocardiogram monitoring against photoelectric isolation noise based on channel attention, comprising the following steps:
[0057] Step 1: Construct an intelligent processing module, including a fractional Fourier transform layer and a one-dimensional convolutional neural network. Train the one-dimensional convolutional neural network using a dataset, as detailed below:
[0058] Step 1.1: Construct an intelligent processing module, including a fractional Fourier transform layer and a one-dimensional convolutional neural network;
[0059] Step 1.2: Construct the dataset and divide it into training, validation, and test sets in a 7:2:1 ratio;
[0060] Step 1.3: Train the one-dimensional convolutional neural network using the weighted cross-entropy loss function and the Adam optimizer until the accuracy on the validation set converges.
[0061] Step 2: Acquire multi-lead ECG signals using an opto-isolated acquisition module;
[0062] Step 3: Perform adaptive length alignment and standardization preprocessing on the acquired signals;
[0063] Step 4: Construct a photoelectric shot noise model and perform data augmentation on the acquired signal, as follows:
[0064] Step 4.1: Construct a photoelectric shot noise model. Treat the normalized ECG signal as simulated light intensity. Based on the Poisson distribution characteristics of photon counting, generate random noise whose variance is proportional to the instantaneous intensity of the signal. Superimpose it onto the original signal. By adjusting the noise intensity coefficient, simulate the acquisition environment with different signal-to-noise ratios.
[0065] Step 4.2: In each training batch, data augmentation is performed on the collected signals with a set probability so that the model learns the essential structural features of the signals, rather than memorizing specific noise patterns.
[0066] Step 5: Input the enhanced signal into the intelligent processing module for feature extraction;
[0067] Step 6: Based on the extracted features, output the ECG diagnosis results.
[0068] The present invention also provides a computer device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the aforementioned channel attention-based anti-optical isolation noise ECG monitoring method by executing the computer instructions.
[0069] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0070] Example
[0071] This embodiment provides a channel attention-based electrocardiogram monitoring system with anti-optical-isolation noise, such as... Figure 1 As shown, it includes an opto-isolated acquisition module, a data preprocessing module, a data enhancement module, an intelligent processing module, and a diagnostic output module;
[0072] The opto-isolated acquisition module is used to acquire multi-lead electrocardiogram signals;
[0073] The data preprocessing module is used to perform adaptive length alignment and standardization processing on the acquired signals;
[0074] The data enhancement module uses a photoelectric shot noise model based on Poisson distribution to perform data enhancement processing on the acquired signal;
[0075] The intelligent processing module assigns adaptive weights to different feature channels of the signal, suppresses noise-related feature channels, and enhances pathology-related feature channels.
[0076] The diagnostic output module is used to output diagnostic results and visual evidence.
[0077] As a specific example, the intelligent processing module includes a fractional Fourier transform layer and a one-dimensional convolutional neural network.
[0078] As a specific example, the fractional Fourier transform layer is placed at the front end of a one-dimensional convolutional neural network to better focus on non-stationary features.
[0079] As a specific example, the one-dimensional convolutional neural network integrates a channel attention module, which employs a Squeeze-and-Excitation module.
[0080] As a specific example, the one-dimensional convolutional neural network employs a one-dimensional residual network.
[0081] As a specific example, the one-dimensional convolutional neural network is trained using a weighted cross-entropy loss function, where the weight of each category of samples in the training set is inversely proportional to the number of samples in that category.
[0082] A channel attention-based method for electrocardiogram monitoring against photoelectric isolation noise includes the following steps:
[0083] Step 1: Construct an intelligent processing module, including a fractional Fourier transform layer and a one-dimensional convolutional neural network. Train the one-dimensional convolutional neural network using a dataset, as detailed below:
[0084] Step 1.1: Build an SE-ResNet1D network based on the PyTorch framework. This network contains four residual blocks, each followed by an SE module, such as... Figure 2 As shown, the SE module obtains channel descriptors through global average pooling, and then generates channel weights through two fully connected layers with a dimensionality reduction ratio of r=16, thereby recalibrating the feature map;
[0085] To capture non-stationary features in electrocardiogram (ECG) signals more precisely, such as QRS waves with linear frequency modulation, this embodiment cascades a differentiable fractional Fourier transform layer at the front end of the SE-ResNet1D network. This layer contains a learnable fractional-order parameter. It is initialized to 1; it performs FFT on the input signal after double phase modulation to calculate the energy distribution in the fractional-order domain; the output is then processed by learnable weights and residually connected to the original time-domain signal to form an enhanced hybrid feature input to the subsequent network;
[0086] This layer provides the network with degrees of freedom to rotate in the time-frequency plane, enabling it to adaptively find the optimal feature representation domain. Experiments show that introducing this layer further enhances the model's ability to capture subtle distortions in LBBB waveforms, such as... Figure 3 As shown.
[0087] Step 1.2: Using 6862 clinical data from partner hospitals, the data were divided into training, validation, and test sets according to patient ID in a 7:2:1 ratio.
[0088] Step 1.3: Train using the weighted cross-entropy loss function, with weights calculated according to the formula. The calculation is performed, where N is the total number of samples and K is the number of categories. In this example, there are four categories: normal, myocardial infarction, left bundle branch block, and others. Let be the number of samples in class c; use the Adam optimizer to train until the accuracy on the validation set converges;
[0089] Step 2: Use the linear optocoupler HCNR201 to construct an isolation amplification front end and acquire 12-lead ECG signals at a sampling rate of 500Hz;
[0090] Step 3: Parse the original HL7 format data stream, use an adaptive length alignment algorithm to unify all signals to a fixed length, such as 5000 points corresponding to 10 seconds, and perform Z-Score normalization.
[0091] Step 4: Construct a photoelectric shot noise model and perform data augmentation on the acquired signal, as follows:
[0092] Step 4.1: Construct a photoelectric shot noise model. Treat the normalized ECG signal as simulated light intensity. Based on the Poisson distribution characteristics of photon counting, generate random noise whose variance is proportional to the instantaneous intensity of the signal. Superimpose it onto the original signal. By adjusting the noise intensity coefficient, simulate the acquisition environment with different signal-to-noise ratios.
[0093] Step 4.2: In each training batch, data augmentation is performed on the collected signals with a set probability so that the model learns the essential structural features of the signals, rather than memorizing specific noise patterns.
[0094] Comparative experiments show that, after adopting this enhancement strategy, the model's accuracy on the low signal-to-noise ratio simulation test set is improved by approximately 1.8% compared to the unenhanced model;
[0095] Step 5: Input the enhanced signal into the intelligent processing module for feature extraction;
[0096] Step 6: Based on the extracted features, output the ECG diagnosis results.
[0097] The diagnostic system software features a visual interactive interface on the computing platform display terminal, mainly divided into three functional areas: 1. Real-time waveform monitoring area: synchronously renders 12-lead ECG waveforms acquired and preprocessed through photoelectric isolation; 2. Intelligent diagnostic conclusion area: displays the pathological classification results (e.g., 'suspected myocardial infarction') and corresponding prediction confidence (e.g., "99.08%)" output by the algorithm in real time; 3. Interpretable visualization area: based on Grad-CAM technology, it overlays an attention heatmap onto the original waveform, highlighting key morphological features (e.g., ST segment elevation or wide and deformed QRS waves) on a high-brightness color (e.g., red) to provide clinicians with intuitive auxiliary diagnostic evidence.
[0098] The evaluation was performed on an independent test set, and the results are as follows: Figure 5 As shown, the system's macro-average AUC reached 0.9430. In the critical LBBB and MI discrimination task, the AUC for LBBB was as high as 0.9791, with a specificity of 98.40%; the AUC for MI was 0.9583, and the system's specificity for normal rhythms also reached 97.88%, demonstrating extremely high security.
[0099] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A channel attention-based electrocardiogram monitoring system with anti-optical-isolation noise, characterized in that, It includes an opto-isolated acquisition module, a data preprocessing module, a data enhancement module, an intelligent processing module, and a diagnostic output module; The opto-isolated acquisition module is used to acquire multi-lead electrocardiogram signals; The data preprocessing module is used to perform adaptive length alignment and standardization processing on the acquired signals; The data enhancement module uses a photoelectric shot noise model based on Poisson distribution to perform data enhancement processing on the acquired signal; The intelligent processing module assigns adaptive weights to different feature channels of the signal, suppresses noise-related feature channels, and enhances pathology-related feature channels. The diagnostic output module is used to output diagnostic results and visual evidence.
2. The electrocardiogram monitoring system based on channel attention and resisting photoelectric isolation noise according to claim 1, characterized in that, The intelligent processing module includes a fractional Fourier transform layer and a one-dimensional convolutional neural network.
3. The electrocardiogram monitoring system based on channel attention and resisting photoelectric isolation noise according to claim 2, characterized in that, The fractional Fourier transform layer is placed at the front end of the one-dimensional convolutional neural network to better focus on non-stationary features.
4. The electrocardiogram monitoring system based on channel attention and resisting photoelectric isolation noise according to claim 2, characterized in that, The one-dimensional convolutional neural network integrates a channel attention module, which employs a Squeeze-and-Excitation module.
5. The electrocardiogram monitoring system based on channel attention and resisting photoelectric isolation noise according to claim 2, characterized in that, The one-dimensional convolutional neural network uses a one-dimensional residual network.
6. The electrocardiogram monitoring system based on channel attention and resisting photoelectric isolation noise according to claim 2, characterized in that, The one-dimensional convolutional neural network is trained using a weighted cross-entropy loss function, and the weight of each class of samples in the training set is inversely proportional to the number of samples in that class.
7. A method for electrocardiogram monitoring based on channel attention to resist photoelectric isolation noise, characterized in that, Includes the following steps: Step 1: Construct an intelligent processing module, including a fractional Fourier transform layer and a one-dimensional convolutional neural network, and train the one-dimensional convolutional neural network using a dataset. Step 2: Acquire multi-lead ECG signals using an opto-isolated acquisition module; Step 3: Perform adaptive length alignment and standardization preprocessing on the acquired signals; Step 4: Construct a photoelectric shot noise model and perform data augmentation on the acquired signals; Step 5: Input the enhanced signal into the intelligent processing module for feature extraction; Step 6: Based on the extracted features, output the ECG diagnosis results.
8. The ECG monitoring method based on channel attention to resist photoelectric isolation noise according to claim 7, characterized in that, The intelligent processing module described in step 1 includes a fractional Fourier transform layer and a one-dimensional convolutional neural network. The one-dimensional convolutional neural network is trained using a dataset, as detailed below: Step 1.1: Construct an intelligent processing module, including a fractional Fourier transform layer and a one-dimensional convolutional neural network; Step 1.2: Construct the dataset and divide it into training, validation, and test sets in a 7:2:1 ratio; Step 1.3: Train the one-dimensional convolutional neural network using the weighted cross-entropy loss function and the Adam optimizer until the accuracy on the validation set converges.
9. The ECG monitoring method based on channel attention to resist photoelectric isolation noise according to claim 7, characterized in that, Step 4 involves constructing a photoelectric shot noise model and performing data augmentation on the acquired signals, as detailed below: Step 4.1: Construct a photoelectric shot noise model. Treat the normalized ECG signal as simulated light intensity. Based on the Poisson distribution characteristics of photon counting, generate random noise whose variance is proportional to the instantaneous intensity of the signal. Superimpose it onto the original signal. By adjusting the noise intensity coefficient, simulate the acquisition environment with different signal-to-noise ratios. Step 4.2: In each training batch, the collected signals are augmented with a set probability to enable the model to learn the essential structural features of the signals.
10. A computer device, characterized in that, include: The system includes a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the electrocardiogram monitoring method based on channel attention based on any one of claims 7 to 9.