Electroencephalogram and electrocardiogram coupling learning-based electrocardiogram sleep staging method

Abstract

The application discloses an electrocardiogram sleep staging method based on electroencephalogram and electrocardiogram coupling learning, improves a UNet basic feature extraction structure, and enables the UNet basic feature extraction structure to extract more rich information of multiple levels and multiple scales. Through a pooling layer, a convolution layer, a normalization layer and an activation layer, basic electrocardiogram and electroencephalogram features are respectively fused, processed, normalized in scale, and supervised coupling learning is adopted by using a mean square error similarity function, so that the features learned by electrocardiogram are as similar as possible to electroencephalogram, and more effective features for electrocardiogram sleep staging are extracted, thereby effectively improving the accuracy and efficiency of (electrocardiogram) sleep staging.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of sleep staging prediction, and particularly relates to an electrocardiogram sleep staging method based on electroencephalogram and electrocardiogram coupling learning. BACKGROUND

[0002] Various physiological parameters of the human body change with different sleep states, and the process of dividing different sleep states according to these vital sign signal characteristics is called sleep staging. Sleep staging is a key link in sleep monitoring, sleep structure analysis and sleep state evaluation.

[0003] Among numerous physiological signals, the electroencephalogram signal is the most obvious signal in the sleep process, but the collection of the electroencephalogram signal is not convenient and the cost is high. The signal collection device is expensive and can only be used in hospitals or professional laboratories, which greatly limits the research of sleep staging and the promotion of home care

[0004] Compared with electroencephalogram detection, electrocardiogram monitoring is more inexpensive and convenient. Many new wearable electrocardiogram devices have also been developed, but in actual use, the accuracy of electrocardiogram sleep staging is low and sleep staging cannot be effectively predicted.

[0005] In summary, the prior art has the problem that it cannot simultaneously consider applicability and accuracy when predicting sleep staging. SUMMARY

[0006] The purpose of the present application is to provide an electrocardiogram sleep staging method based on electroencephalogram and electrocardiogram coupling learning to solve the problem that the sleep staging prediction method in the prior art cannot simultaneously consider applicability and accuracy.

[0007] In order to solve the above technical problems, the present application adopts the following technical solutions to achieve it:

[0008] An electrocardiogram sleep staging method based on electroencephalogram and electrocardiogram coupling learning comprises the following steps:

[0009] Step 1, collect a data set of electrocardiogram signals and electroencephalogram signals, delete the meaningless segments in the data set of electrocardiogram signals and electroencephalogram signals, and then resample to obtain a resampled data set of electrocardiogram signals and electroencephalogram signals;

[0010] Signal staging is performed on the resampled data set of electrocardiogram signals and electroencephalogram signals to obtain an initial data set of electrocardiogram signals and electroencephalogram signals;

[0011] Step 2, input the initial data set of electrocardiogram signals and electroencephalogram signals into an electrocardiogram sleep staging model, and use the basic feature extraction structure of the electrocardiogram sleep staging model to extract electrocardiogram basic features and electroencephalogram basic features from the initial data set of electrocardiogram signals and electroencephalogram signals;

[0012] Step 3, using the electrocardiogram sleep staging model, the electrocardiogram basic features and the electroencephalogram basic features are coupled and learned using a coupling learning structure, and similar features of the electrocardiogram basic features and the electroencephalogram basic features are extracted;

[0013] Step 4, using the similar features and using a loss function, the electrocardiogram sleep staging model is supervised and trained, and the trained electrocardiogram sleep staging model is obtained.

[0014] Step 5, the electrocardiogram signal to be predicted is input into the trained electrocardiogram sleep staging model, and a prediction result is obtained.

[0015] The present application also has the following features:

[0016] Further, in step 1,

[0017] The segments without practical significance include the wake-up segments monitored before sleep and the segments without sleep staging labels;

[0018] The resampled electrocardiogram signal and the electroencephalogram signal data set are signal staged according to the AASM Institute standard.

[0019] Further, step 2 includes the following steps:

[0020] Step 21, the initial electrocardiogram signal and the electroencephalogram signal data set are input into the basic feature extraction structure of the electrocardiogram sleep staging model; at the same time, the two layers of convolution of the basic feature extraction structure of the electrocardiogram sleep staging model are pooled and normalized;

[0021] Step 22, using the attention structure for decoding, the data in the initial electrocardiogram signal and the electroencephalogram signal data set are processed for features;

[0022] Step 23, using the skip connection structure to connect the information between different layers in the initial electrocardiogram signal and the electroencephalogram signal data set;

[0023] Step 24, using the normalized basic feature extraction structure for feature extraction, the electrocardiogram basic features and the electroencephalogram basic features are obtained.

[0024] Further, the basic feature extraction structure is UNet.

[0025] Further, in step 21, after the two layers of convolution of the basic feature extraction structure are pooled and normalized, the first layer of convolution uses a 1*5 convolution kernel; the second layer of convolution uses a 1*1 convolution kernel.

[0026] In step 24, when the basic feature extraction structure is used for feature extraction, a 1*1, 1*3 and 1*5 three different size convolution kernels are used.

[0027] Further, step 3 comprises the following sub-steps:

[0028] Step 31, pool the electrocardio base feature map of the electrocardio base feature and the electroencephalogram base feature map of the electroencephalogram base feature;

[0029] The pooled electrocardio base feature map and electroencephalogram base feature map are fused and normalized in scale;

[0030] Step 32, optimize the base feature extraction structure using an activation function;

[0031] Step 33, using the electrocardio sleep staging model, the electrocardio base feature map and electroencephalogram base feature map after fusion and scale normalization are coupled learning based on the coupling learning structure to obtain similar features of the electrocardio base feature and electroencephalogram base feature.

[0032] Further, in step 31, a pooling layer with a size of (1, 10) is used to pool the electrocardio base feature map of the electrocardio base feature and the electroencephalogram base feature map of the electroencephalogram base feature;

[0033] The pooled electrocardio base feature map and electroencephalogram base feature map are fused and normalized in scale using a 1*1 convolution and a normalization layer;

[0034] In step 32, the activation function uses Leaky ReLU;

[0035] In step 33, during the coupling learning process of the electrocardio sleep staging model on the electrocardio base feature map and electroencephalogram base feature map after fusion and scale normalization, mean square error similarity function is used to supervise the coupling learning, and the mean square error similarity function is as follows:

[0036]

[0037] Wherein, m represents the number of sleep stages in a batch;

[0038] f eeg represents the features extracted from the electroencephalogram signal in each sleep stage time period;

[0039] f ecg represents the features extracted from the electrocardio signal in each sleep stage time period.

[0040] Further, step 4 comprises the following sub-steps:

[0041] Step 41, using the electrocardio sleep staging model after coupling learning obtained in step 33, electrocardio features and electroencephalogram features are extracted from the initial electrocardio signal and electroencephalogram signal data set;

[0042] Step 42, the extracted electrocardio features and electroencephalogram features are sent to the full connection layer of the electrocardio sleep staging model after coupling learning to obtain the prediction data result of electrocardio sleep staging;

[0043] Step 43, the prediction data result is compared with the true label of the data of the initial electrocardio signal and electroencephalogram signal, and is substituted into the loss function to calculate the prediction error value, and the iteration training is performed to obtain the electrocardio sleep staging model after training.

[0044] Further, the loss function in step 43 adopts CrossEntropyLoss.

[0045] Compared with the prior art, the present application has the following technical effects:

[0046] (I) The electrocardio sleep staging method based on electroencephalogram and electrocardio coupling learning of the present application improves the UNet basic feature extraction structure, so that more rich information of multi-level and multi-scale can be extracted. The basic electrocardio and electroencephalogram features are respectively fused, normalized in scale, and supervised coupling learning is performed by using the mean square error similarity function, so that the features learned by electrocardio are as similar as possible to electroencephalogram, thereby more effective features for electrocardio sleep staging are extracted, and the accuracy and applicability of sleep staging are effectively improved.

[0047] (II) The electrocardio sleep staging method based on electroencephalogram and electrocardio coupling learning of the present application can be applied to an electrocardio detection device. There are many wearable electrocardio devices in the prior art, and the method of the present application can produce a synergistic effect, which is helpful for the further development of sleep research and lays a good foundation for the promotion of home sleep devices.

[0048] (III) The electrocardio sleep staging method based on electroencephalogram and electrocardio coupling learning of the present application simplifies the sleep staging process, without manual feature extraction and complex feature engineering. Instead, the deep learning network can comprehensively capture the spatio-temporal information and semantic features in the electroencephalogram and electrocardio signal data, automatically learn and extract the electrocardio signal features, and simplify the sleep staging process. Experiments show that the classification method of the present application has a significant improvement in classification accuracy in the electrocardio sleep staging task compared with other existing models, and obtains a high sleep staging accuracy in the electrocardio sleep staging task, which is suitable for large-scale use and promotion in industry. DETAILED DESCRIPTION

[0049] It should be noted that all methods in the present application use known methods in the prior art unless otherwise specified.

[0050] The following gives specific embodiments of the present application, and it should be noted that the present application is not limited to the following specific embodiments, and any equivalent transformations made on the basis of the technical solutions of the present application fall within the protection scope of the present application.

[0051] A method for electrocardiogram sleep staging based on electroencephalogram and electrocardiogram coupling learning, comprising the following steps:

[0052] Step 1, collect the data set of electrocardiogram signals and electroencephalogram signals, delete the meaningless segments in the data set of electrocardiogram signals and electroencephalogram signals, and then resample to obtain the resampled data set of electrocardiogram signals and electroencephalogram signals;

[0053] Signal staging is performed on the resampled data set of electrocardiogram signals and electroencephalogram signals to obtain an initial data set of electrocardiogram signals and electroencephalogram signals;

[0054] Step 2, using a UNet basic feature extraction structure, extract electrocardiogram basic features and electroencephalogram basic features from the initial data set of electrocardiogram signals and electroencephalogram signals;

[0055] Step 3, based on the coupling learning structure, using the electrocardiogram sleep staging model, coupling learning of the electrocardiogram basic features and the electroencephalogram basic features is performed, and then similar features of the electrocardiogram basic features and the electroencephalogram basic features are extracted; the purpose of this step is to enable the electrocardiogram signal to have the accuracy of the electroencephalogram signal for sleep staging;

[0056] Step 4, using the similar features and using the loss function, the electrocardiogram sleep staging model is supervised trained to obtain the UNet basic feature extraction structure after training.

[0057] Step 5, input the electrocardiogram signal to be predicted into the trained electrocardiogram sleep staging model to obtain a prediction result.

[0058] Further, the meaningless segments in step 1 include the wake-up segments monitored in the pre-sleep stage and the segments without sleep staging labels;

[0059] When signal staging is performed on the resampled data set of electrocardiogram signals and electroencephalogram signals, signal staging is performed according to the AASM Institute standard.

[0060] Further, step 2 includes the following sub-steps:

[0061] Step 21, input the initial data set of electrocardiogram signals and electroencephalogram signals into the basic feature extraction structure of the electrocardiogram sleep staging model; at the same time, pool and normalize the two layers of convolution of the basic feature extraction structure of the electrocardiogram sleep staging model;

[0062] Step 22, decoding using attention structure, feature processing is performed on the data in the data set of the initial electrocardiogram signal and electroencephalogram signal; it should be noted that the attention decoding structure mainly includes Squeeze and Excitation operations.

[0063] The Squeeze operation is to compress the spatial dimension of the input feature map through global pooling, thereby converting it into a channel dimension feature vector. This feature vector represents the global feature importance of each channel. This operation reduces the dimension of the feature map while extracting the global representation on the current feature.

[0064] The Excitation operation uses a fully connected neural network to learn the relationship between channels and generates a channel attention weight vector. This weight vector is normalized by the activation function Sigmoid to ensure that the weight is between 0 and 1, so that the attention weight of each channel represents its importance in the entire feature map. By adaptively adjusting the channel weight in this way, relevant feature channels can be enhanced and irrelevant feature channels can be suppressed, thereby improving the expression ability of the UNet basic feature extraction structure.

[0065] Step 23, using a skip connection structure to connect the information between different layers in the data set of the initial electrocardiogram signal and electroencephalogram signal; the skip connection structure connects the information between different layers in the data set of the initial electrocardiogram signal and electroencephalogram signal. The purpose of this step is to promote the flow and sharing of information, so that the basic feature extraction structure can more easily learn residual information, which helps to alleviate the problems of gradient vanishing and gradient explosion, and speeds up the training process. Through the skip connection, the shallow features can be directly passed to the deep layers, so that the network can better utilize the correlation between low-level and high-level features, and improve the learning ability of complex patterns and abstract features.

[0066] Step 24, using the normalized basic feature extraction structure to extract features to obtain electrocardiogram basic features and electroencephalogram basic features.

[0067] Furthermore, in step 21, after the two layers of convolution of the basic feature extraction structure are pooled and normalized, the first layer of convolution uses a 1*5 convolution kernel; the second layer of convolution uses a 1*1 convolution kernel; the purpose of this step is to compress the expanded feature dimension, remove features with low task relevance, and retain the most important feature information, to reduce the amount of calculation and enhance the robustness of the UNet basic feature extraction structure.

[0068] Normalizing the data allows the basic feature extraction structure to learn scaling and offset parameters, thereby improving stability and convergence speed.

[0069] When feature extraction is performed using the basic feature extraction structure, three different sizes of convolution kernels, 1*1, 1*3 and 1*5, are used.

[0070] Further, step 3 comprises the following sub-steps:

[0071] Step 31, pooling the electrocardiogram basic feature map of the electrocardiogram basic feature and the electroencephalogram basic feature map of the electroencephalogram basic feature;

[0072] The pooled electrocardiogram basic feature map and electroencephalogram basic feature map are fused and normalized in scale;

[0073] Step 32, using an activation function to optimize the basic feature extraction structure;

[0074] Step 33, using the electrocardiogram sleep staging model, coupling learning is carried out based on the coupling learning structure for the fused and normalized electrocardiogram basic feature map and electroencephalogram basic feature map to obtain similar features of the electrocardiogram basic feature and the electroencephalogram basic feature.

[0075] Further, in step 31, a pooling layer with a size of (1, 10) is used to pool the electrocardiogram basic feature map and electroencephalogram basic feature map of the electrocardiogram basic feature and the electroencephalogram basic feature;

[0076] The pooled electrocardiogram basic feature map and electroencephalogram basic feature map are fused and normalized in scale using a 1*1 convolution and a normalization layer;

[0077] The electrocardiogram and electroencephalogram basic features are fused and normalized in scale respectively, which aims to eliminate the scale difference between the data, improve the consistency of data processing and analysis, enhance the comprehensiveness of the features, reveal the correlation between the data, provide more representative features, and thus improve the accuracy and efficiency of data analysis and modeling.

[0078] Since the basic electrocardiogram and electroencephalogram feature maps extracted after the UNet basic feature extraction structure optimized in step 2 have a shape of (Batchsize, 32, 1, 60000), in order to reduce the amount of calculation and reduce the influence between modalities due to the different unit scales of the basic features, a pooling layer with a size of (1, 10) is used to reduce the spatial dimension of the feature map, reduce the computational complexity, and help the feature to have translational invariance. Secondly, a 1*1 convolution and a normalization layer are used to fuse the features, normalize the scale, help to improve the stability of training, accelerate convergence, reduce overfitting, and help to improve performance.

[0079] In step 32, the activation function uses Leaky ReLU;

[0080] The activation layer can learn more complex and abstract feature representations, improve the model's adaptability to complex data, and better fit the variable sleep data, thereby improving its ability to capture and express data features.

[0081] The Leaky ReLU can solve the problem of neuron death that may occur with the ReLU activation function, and by introducing a non-zero slope, the Leaky ReLU allows negative inputs to have a gradient. It also helps to alleviate the gradient vanishing problem and improve the training effect and expression ability of the neural network.

[0082] In step 33, during the coupling learning process of the fused and scale-normalized electrocardiogram and electroencephalogram base feature maps, the mean square error similarity function is used to supervise the coupling learning, and the mean square error similarity function is as follows:

[0083]

[0084] Where m represents the number of sleep stages in a batch;

[0085] f eeg represents the features extracted from the electroencephalogram signal in each sleep stage time period;

[0086] f ecg represents the features extracted from the electroencephalogram signal in each sleep stage time period.

[0087] It should be noted that the mean square error similarity function is used to supervise the coupling learning, so that the electroencephalogram signal assists the electrocardiogram signal training, and the features learned by the electrocardiogram signal and the features learned by the electroencephalogram signal are as similar as possible, so that more effective features for electrocardiogram sleep staging are extracted, thereby further improving the accuracy of sleep staging. Finally, only electrocardiogram is used to further improve the efficiency of sleep staging.

[0088] Mean square error (MSE) is a commonly used loss function and evaluation index, which is used to measure the difference between the predicted value and the true value. It calculates the average squared distance between the predicted value and the true value to evaluate the accuracy of the model. The MSE is used to compare the differences between the electroencephalogram (EEG) and electrocardiogram (ECG) signals. By minimizing the MSE loss function, the features extracted from the EEG and ECG can be made more similar. This helps to reduce the modal-specific differences between EEG and ECG, and promotes the electrocardiogram signal to learn more similar features to the electroencephalogram signal, so that the electrocardiogram signal extracts more effective feature information for sleep staging.

[0089] Further, step 4 comprises the following sub-steps:

[0090] Step 41, using the coupled learning electrocardiosleep staging model obtained in step 33 to extract electrocardio features and electroencephalogram features from the initial electrocardio signal and electroencephalogram signal data set;

[0091] Step 42, the extracted electrocardio features and electroencephalogram features are sent to the full connection layer of the coupled learning electrocardiosleep staging model to obtain the prediction data result of electrocardiosleep staging;

[0092] Step 43, the prediction data result is compared with the true label of the initial electrocardio signal and electroencephalogram signal data, and the prediction error value is calculated by substituting it into the loss function, and the training is iterated to obtain the trained electrocardiosleep staging model.

[0093] Further, the loss function in step 43 adopts CrossEntropyLoss.

[0094] The CrossEntropyLoss loss function has a built-in Softmax layer, which records and feeds back the error value to the model in real time. Through multiple rounds of data iteration training, the weights and biases of the network nodes are constantly optimized and updated, so that the final classification result is closer to the true label. After completing the above parameter training, the test data is responded to the decision.

[0095] In order to verify the feasibility and effectiveness of the scheme, the accuracy and MF1 score are used to judge the result of the electrocardiosleep staging model.

[0096] The main task of the scheme is W, NREM, REM three-stage and W, N1, N2, N3, REM five-stage task. The evaluation index includes accuracy and F1 score.

[0097] The accuracy is the ratio of the number of correct sleep stages predicted by the model to the total sleep stage data,

[0098] The F1 score is an index used to measure the accuracy of the classification model, which takes into account both the accuracy and recall of the classification model. It can be regarded as a weighted average of the accuracy and recall of the model.

[0099] MF1 score can help evaluate the overall performance of the model in multi-class classification problems without being affected by the imbalance of class sample size.

[0100] MASS comes from the Montreal Sleep Study Archives, which has 5 subsets SS1-SS5. These subsets are arranged in order of their research and acquisition protocols.

[0101] Table 1 is a performance index comparison table of the present scheme and other methods, which shows the 31-fold cross-validation classification results of the MASS-SS3 dataset, wherein the three-stage results and the five-stage results reach 79.61% and 68.28%, respectively, which are better than those of other comparison models, and the present scheme has a significant advantage in electrocardio sleep staging.

[0102] Table 1: Performance index comparison table of the present scheme and other methods

[0103] Method of choice Staging ACC MF1 TinySleepNet 3 71.70 50 SalientSleepNet 3 70.86 52.74 TransSleep 3 71.06 49.47 The present solution 3 79.61 70.68 TinySleepNet 5 46 31.8 SalientSleepNet 5 56.17 43.79 TransSleep 5 51.64 36.51 The present solution 5 68.28 58.52

[0104] Among them, TinySleepNet, SalientSleepNet and TransSleep are the names of existing methods known in the prior art, and those skilled in the art can know the methods corresponding to the above names.

[0105] Table 2 and Table 3 are respectively a three-stage performance index comparison table before and after coupling learning of the present scheme and a five-stage performance index comparison table before and after coupling learning of the present scheme; and the performance comparison in each staging task before and after coupling of the present model using electroencephalogram and electrocardio signals is respectively shown.

[0106] The results show that the performance after coupling is better than that before coupling, and the accuracy of electrocardio sleep staging is effectively improved under the auxiliary training of electroencephalogram.

[0107] Table 2: Three-stage performance index comparison table before and after coupling learning of the present scheme

[0108]

[0109] Table 3: Five-stage performance index comparison table before and after coupling learning of the present scheme

[0110]

Claims

1. A method for ECG sleep staging based on coupled learning of electroencephalography (EEG) and electrocardiography (ECG), characterized in that, Includes the following steps: Step 1: Collect the datasets of electrocardiogram (ECG) and electroencephalogram (EEG) signals. After deleting meaningless segments from the ECG and EEG datasets, resample the data to obtain the resampled ECG and EEG datasets. The resampled ECG and EEG signal datasets were segmented to obtain the initial ECG and EEG signal datasets. Step 2: Input the initial ECG and EEG signal datasets into the ECG sleep staging model, and use the basic feature extraction structure of the ECG sleep staging model to extract basic ECG and EEG features from the initial ECG and EEG signal datasets. Step 3: Use the ECG sleep staging model to perform coupled learning on the basic ECG features and basic EEG features using a coupled learning structure, and extract similar features between the basic ECG features and basic EEG features. Step 4: Use similar features and a loss function to supervise the training of the ECG sleep staging model to obtain the trained ECG sleep staging model. Step 5: Input the ECG signal to be predicted into the trained ECG sleep staging model to obtain the prediction result.

2. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 1, characterized in that, In step 1: Segments that are not of practical significance include waking segments from pre-sleep monitoring and segments without sleep stage labels; The resampled ECG and EEG datasets were segmented according to the AASM standards.

3. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 1, characterized in that, Step 2 includes the following sub-steps: Step 21: Input the initial ECG and EEG signal datasets into the basic feature extraction structure of the ECG sleep staging model; simultaneously, perform pooling and normalization on the two convolutional layers of the basic feature extraction structure of the ECG sleep staging model. Step 22: Decoding is performed using an attention structure to perform feature processing on the data in the initial ECG and EEG signal datasets; Step 23: Use a skip connection structure to connect information between different layers in the initial ECG and EEG signal datasets; Step 24: Use the normalized basic feature extraction structure to extract features and obtain basic ECG and EEG features.

4. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 3, characterized in that, The basic feature extraction structure is UNet.

5. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 4, characterized in that, In step 24, when using the basic feature extraction structure for feature extraction, three different scales of convolution kernels are used: 1*1, 1*3, and 1*5.

6. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 1, characterized in that, Step 3 includes the following sub-steps: Step 31: Pooled ECG baseline feature map, pooled EEG baseline feature map; The pooled ECG and EEG baseline feature maps were fused and scaled. Step 32: Optimize the basic feature extraction structure using an activation function; Step 33: Using the ECG sleep staging model, perform coupling learning on the fused and scale-normalized ECG and EEG basic feature maps based on the coupling learning structure to obtain similar features between the ECG and EEG basic features.

7. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 6, characterized in that... ； In step 31, a pooling layer of size (1, 10) is used to pool the basic ECG feature map of the basic ECG feature and the basic EEG feature map of the basic EEG feature, respectively. A 1*1 convolutional and normalization layer is used to fuse and scale-normalize the pooled ECG and EEG baseline feature maps. In step 32, the activation function used is Leaky ReLU; In step 33, during the coupled learning process of the ECG and EEG baseline feature maps after fusion processing and scale normalization using the ECG sleep staging model, the mean squared error similarity function is used to supervise the coupled learning. The mean squared error similarity function is as follows: Where m represents the number of sleep stages in a batch; This represents the features extracted from EEG signals for each sleep stage. This represents the features extracted from the electrocardiogram signal for each sleep stage.

8. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 7, characterized in that, Step 4 includes the following sub-steps: Step 41: Using the coupled learning ECG sleep staging model obtained in Step 33, extract ECG features and EEG features from the initial ECG and EEG signal datasets. Step 42: The extracted ECG and EEG features are fed into the fully connected layer of the coupled learning ECG sleep staging model to obtain the predicted data results of ECG sleep staging. Step 43: Compare the predicted data results with the true labels of the initial ECG and EEG signals, substitute them into the loss function to calculate the prediction error value, perform iterative training, and obtain the trained ECG sleep staging model.

9. The electrocardiogram sleep staging method based on EEG and ECG coupling learning as described in claim 8, characterized in that, The loss function used in step 43 is CrossEntropyLoss.

Images