Consciousness state monitoring method and system based on graph attention and temporal convolution

Abstract

The application provides a consciousness state monitoring method and system based on graph attention and time convolution, relates to the technical field of electroencephalogram signal processing, and specifically includes the following steps: acquiring sampled electroencephalogram data and segmenting the electroencephalogram data; taking each segment of the electroencephalogram data as a node, extracting entropy features and power spectrum features of each segment of the electroencephalogram data as node features, and then constructing a graph based on the correlation between the nodes; inputting the graph into a trained consciousness state monitoring network to obtain a consciousness state index representing a consciousness state; the application extracts mixed features of the electroencephalogram data through graph attention, and performs feature enhancement on the mixed features by using time dependence and time convolution, so as to find the relationship between the electroencephalogram data and the consciousness state, and solve the problems of time lag and individual difference existing in the prior monitoring method.

Description

Technical Field

[0001] This invention belongs to the field of EEG signal processing technology, and particularly relates to a method and system for monitoring consciousness state based on graph attention and temporal convolution. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] The state of consciousness refers to the level of activity of the patient's brain and can be used to assess the patient's anesthesia status. Both excessively deep and insufficient anesthesia can cause mental and physical harm to the patient during surgery. Insufficient anesthesia may lead to increased motor responses, making surgical procedures more difficult and increasing surgical risks. Excessively deep anesthesia may suppress the patient's respiratory center, slowing the respiratory rate or even causing respiratory arrest, leading to hypoxia or even asphyxia. It may also cause circulatory system depression, reducing cardiac contractility and blood output, triggering serious complications such as arrhythmia and hypotension, and in severe cases, even endangering the patient's life.

[0004] With the increasing precision of surgical procedures and the growing emphasis on medical costs, the control of the depth of anesthesia during surgery requires greater accuracy. However, clinical physiological characteristics, as the primary method for assessing the depth of anesthesia, suffer from drawbacks such as difficulty in quantification and ambiguity, failing to meet current clinical anesthesia requirements. Therefore, establishing an effective and reliable method for monitoring the state of consciousness is of great significance.

[0005] Electroencephalogram (EEG) signals are a comprehensive representation of the electrophysiological activity of neurons in the cerebral cortex, containing rich information about brain activity and reflecting the subject's state of consciousness. Because EEG provides high temporal resolution information about brain bioelectrical activity, it can accurately and in real-time reflect a patient's brain function and psychological state. Therefore, it is considered a potential objective biomarker for diagnosing mental illnesses and cognitive impairments and is widely used in anesthesia monitoring.

[0006] In recent years, electroencephalography (EEG) has become the preferred technique for measuring consciousness, and significant progress has been made in EEG-based methods for monitoring consciousness. Among these, the bispectral index (BIS), which uses dimensionless numbers to reflect the degree of brain inhibition and thus assess consciousness, has been widely used clinically. However, while BIS has proven to have high sensitivity and a certain degree of accuracy, studies have also pointed out that it suffers from time lag and individual variability. Furthermore, due to the high time dependence of EEG signals, current mainstream EEG-based methods for monitoring consciousness cannot accurately capture dynamic changes in EEG, resulting in overall poor monitoring performance. Summary of the Invention

[0007] To overcome the shortcomings of the prior art, this invention provides a method and system for monitoring consciousness state based on graph attention and temporal convolution. By extracting mixed features from EEG data through graph attention, and enhancing the mixed features using time dependence and temporal convolution, the relationship between EEG data and consciousness state can be found, thus solving the problems of time lag and individual differences in existing monitoring methods.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] The first aspect of this invention provides a method for monitoring conscious states based on graph attention and temporal convolution.

[0010] Consciousness state monitoring methods based on graph attention and temporal convolution include:

[0011] Acquire and segment the electroencephalogram (EEG) data of the individuals to be monitored;

[0012] Using each segment of EEG data as a node, the entropy and power spectrum features of each segment of EEG data are extracted as node features, and then a graph is constructed based on the correlation between nodes.

[0013] The graph is input into the trained consciousness state monitoring network to obtain the consciousness state index, which represents the consciousness state.

[0014] The consciousness state monitoring network extracts high-dimensional temporal features from the graph through graph attention, fuses the high-dimensional temporal features with node features, enhances the fused hybrid features through temporal convolution, and calculates the consciousness state index using the enhanced hybrid features.

[0015] Furthermore, the acquisition of the EEG data specifically involves:

[0016] Perform single-channel EEG data sampling;

[0017] Segmented according to preset time intervals.

[0018] Furthermore, it also includes preprocessing the segmented EEG data, including bandpass filtering, baseline drift removal, and notch filtering.

[0019] Furthermore, the entropy features include four types: wavelet entropy, permutation entropy, sample entropy, and differential entropy; these four entropy features are extracted from each segment of EEG data.

[0020] The power spectral characteristics are the power spectral densities of the four bands of EEG data: Delta, Theta, Alpha, and Beta.

[0021] Furthermore, the construction graph is specifically as follows:

[0022] The correlation between each EEG data segment is calculated as the adjacency matrix, and the entropy feature and power spectrum feature of each EEG data segment are used as node features to construct a node feature matrix, resulting in a graph composed of the adjacency matrix and the node feature matrix.

[0023] Furthermore, the consciousness state monitoring network includes a deep feature extraction module, a feature fusion layer, and a feature enhancement module;

[0024] The deep feature extraction module, based on graph attention, is used to extract high-dimensional temporal features from the graph;

[0025] The feature fusion layer is used to fuse high-dimensional temporal features and node features to obtain hybrid features;

[0026] The feature enhancement module is used to enhance the features of the mixture, including time-step-based encoding and time-series convolution-based enhancement.

[0027] Furthermore, the time-step-based encoding utilizes a time encoder to encode the time steps of the mixed features based on the high temporal dependence of the EEG data.

[0028] A second aspect of the present invention provides a consciousness state monitoring system based on graph attention and temporal convolution.

[0029] A consciousness state monitoring system based on graph attention and temporal convolution includes a data acquisition module, a graph construction module, and an exponent calculation module.

[0030] The data acquisition module is configured to acquire and segment the EEG data of the person to be monitored.

[0031] The graph construction module is configured to: take each segment of EEG data as a node, extract the entropy features and power spectrum features of each segment of EEG data as node features, and then construct a graph based on the correlation between nodes.

[0032] The index calculation module is configured to input the graph into the trained consciousness state monitoring network to obtain a consciousness state index that represents the consciousness state.

[0033] The consciousness state monitoring network extracts high-dimensional temporal features from the graph through graph attention, fuses the high-dimensional temporal features with node features, enhances the fused hybrid features through temporal convolution, and calculates the consciousness state index using the enhanced hybrid features.

[0034] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of the consciousness state monitoring method based on graph attention and temporal convolution as described in the first aspect of the present invention.

[0035] A fourth aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the consciousness state monitoring method based on graph attention and temporal convolution as described in the first aspect of the present invention.

[0036] The above one or more technical solutions have the following beneficial effects:

[0037] This invention extracts four entropy features—wavelet entropy, permutation entropy, sample entropy, and differential entropy—from EEG signals, along with the power spectral density of four bands: Delta, Theta, Alpha, and Beta. Graph attention is used to perform deep temporal feature extraction on the extracted features. Then, the high-dimensional temporal features are mixed with the extracted features to form a set of hybrid features with multiple scales. This set of hybrid features is fed into a time encoder to mark each time step. The encoded hybrid feature set is then fed into a TCN feature enhancer to enhance the features. The enhanced hybrid features are used to predict the consciousness state index. This invention solves the problems of lag and individual differences in existing methods, improves the real-time performance and accuracy of consciousness state monitoring, and achieves better consciousness state monitoring results.

[0038] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0039] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0040] Figure 1 This is a flowchart of the method in the first embodiment.

[0041] Figure 2 This is a structural diagram of the consciousness state monitoring network in the first embodiment. Detailed Implementation

[0042] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0043] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0044] Example 1

[0045] One embodiment of this disclosure provides a method for monitoring states of consciousness based on graph attention and temporal convolution, such as... Figure 1 As shown, it includes the following steps:

[0046] Step S1: Acquire the EEG data of the person to be monitored and segment it;

[0047] S101. Perform single-channel EEG data sampling:

[0048] EEG data consists of one-dimensional data from a single channel, with a sampling frequency of 100Hz.

[0049] S102. Segment according to a preset time interval:

[0050] The EEG data was segmented, with each 5-second segment representing a separate EEG data point.

[0051] The time interval used for segmentation is related to the labeled data used for training, which will be explained in detail in the following section on network training datasets.

[0052] S103. Preprocess the segmented EEG data, including bandpass filtering, baseline drift removal, and notch filtering:

[0053] Noise and power frequency interference are removed using a sixth-order Butterworth bandpass filter (0.8-49Hz) and a 50Hz notch filter; then, a de-stressing operation is used to remove the baseline drift of the signal, specifically:

[0054] 1) Since the main frequency range of EEG signals is 0-50Hz, and since the frequencies of 0Hz and 50Hz are subject to power frequency or environmental interference, a 0.8-49Hz sixth-order Butterworth filter is used to bandpass filter the signal.

[0055] 2) Use detrending operations to eliminate baseline drift of EEG signals. In this embodiment, the smooth function is used for detrending operations.

[0056] 3) Since power frequency interference is difficult to remove, the Butterworth filter cannot completely filter out 50Hz power frequency interference. Therefore, a 50Hz notch filter was introduced to remove power frequency interference to the greatest extent possible.

[0057] Step S2: Using each EEG data segment as a node, extract the entropy and power spectrum features of each EEG data segment as node features, and then construct a graph based on the correlation between nodes, specifically:

[0058] Step S201: Extract the entropy features and power spectrum features of each EEG data segment:

[0059] Specifically, four entropy features—wavelet entropy, permutation entropy, sample entropy, and differential entropy—are extracted from each segment of preprocessed EEG data, along with the power spectral density (PSD) of the four bands: Delta, Theta, Alpha, and Beta.

[0060] Wavelet entropy uses wavelet transform to decompose EEG signals into subbands of different frequencies, then calculates the entropy value of each subband to measure the signal's irregularity or complexity. Permutation entropy, by calculating the permutation entropy of patterns in an EEG signal sequence, quantifies the uncertainty and complexity of its time series, helping to identify changes in brain activity states. Sample entropy is an indicator of the randomness or complexity of a time series, used to assess the irregularity and complexity of EEG activity. Differential entropy is an indicator of signal complexity; based on the concept of entropy, it estimates the uncertainty of EEG signals by calculating their probability distribution, reflecting the complexity and variability of EEG activity. Power spectrum analysis is a very common analytical method in EEG analysis, typically used to analyze the energy changes of various frequency components in EEG signals.

[0061] The differential entropy is divided into six frequency bands: 1-4Hz, 4-8Hz, 8-13Hz, 13-30Hz, 30-50Hz, and 1-49Hz, resulting in six differential entropy features. Furthermore, according to the calculation formula, a total of 61 features can be obtained, including 49 wavelet entropy features, 1 permutation entropy feature, 1 sample entropy feature, and 4 power spectrum features.

[0062] The specific formula for calculating wavelet entropy is as follows:

[0063]

[0064] Where k = 1, 2, ..., N, represents the wavelet decomposition level, ED k E represents the energy at the k-th resolution level. total This represents the total energy of the signal.

[0065] The specific formula for calculating permutation entropy is:

[0066]

[0067] Where m represents the dimension of the vector, t represents the time delay, P(i) represents the probability of the i-th permutation, and N represents the total number of sequences.

[0068] The specific formula for calculating differential entropy is:

[0069]

[0070] Wherein, the input sequence x represents EEG data, which approximately follows a Gaussian distribution N(μ,σ). 2 ).

[0071] The specific formula for calculating the power spectrum is:

[0072]

[0073] Where x represents the input signal, N is the length of the input signal, and ω is the frequency of the input signal.

[0074] The specific formula for calculating sample entropy is:

[0075] Suppose a sequence x n = [x1,x2,...x n The calculation process for sample entropy is as follows:

[0076] 1) For x n For example, one of its m-dimensional vectors can be represented as x. i =[x(i),x(i+1),...x(i+m-1)], m <n

[0077] 2) Calculate two m-dimensional vectors x i and x j The distance between them is denoted as d(x) i ,x j ):

[0078] d(x i ,x j )=max(|x(i+k)-x(j+k)|)

[0079] 3)) Calculate to satisfy d(x i ,x j )<r, and record the number as n i (m,r), then calculate P i (m,r):

[0080]

[0081] 4) Calculate the average value of P i (m,r), and denote it as B(m,r):

[0082]

[0083] 5) Calculate B(m+1,r) according to the previous four steps, and finally obtain the sample entropy of x n :

[0084]

[0085] Step S202: Construct a graph:

[0086] Taking each segment of electroencephalogram data as a node, calculating the correlation between each segment of electroencephalogram data as an adjacency matrix, taking these 61 features as the node features of each segment of electroencephalogram data, constructing a node feature matrix, and forming a graph with 50 segments of data, so as to obtain a graph composed of a 50×54 node feature matrix and a 50×50 adjacency matrix.

[0087] The core of the method designed in this embodiment lies in TCN. In this embodiment, the receptive field of TCN is 63, and the number of nodes of the graph is linked to the time step, which theoretically should not exceed 63. After many experiments, 50 segments of data are the optimal result that balances the real-time performance of monitoring and the accuracy of the model after many experiments.

[0088] Step S3: Input the graph into the trained consciousness state monitoring network to obtain a consciousness state index representing the consciousness state, where the consciousness state index is a value between 0 and 1, which can reflect the patient's anesthesia depth, and a larger value represents a deeper anesthesia depth.

[0089] Wherein, the consciousness state monitoring network extracts high-dimensional temporal features of the graph through graph attention, performs feature fusion on the high-dimensional temporal features and node features, performs feature enhancement on the fused mixed features through temporal convolution, and calculates the consciousness state index by using the enhanced mixed features.

[0090] As Figure 2As shown, the consciousness state monitoring network includes a deep feature extraction module, a feature fusion layer, and a feature enhancement module. The deep feature extraction module, based on graph attention, is used to extract high-dimensional temporal features from the graph. The feature fusion layer is used to fuse high-dimensional temporal features and node features to obtain hybrid features. The feature enhancement module is used to enhance the hybrid features, including time-step-based encoding and temporal convolution-based enhancement. The time-step-based encoding uses a temporal encoder to encode the time steps of the hybrid features based on the high temporal dependence of EEG data. The temporal convolution-based enhancement uses a temporal convolutional network to further enhance the hybrid features.

[0091] The deep feature extraction module based on graph attention (GAT) consists of four graph attention layers and one normalization layer. The input of this module is the graph generated in step S2, through which three deep temporal features can be extracted.

[0092] The feature fusion layer, following the deep feature extraction module, concatenates high-dimensional temporal features and node features to obtain hybrid features.

[0093] The feature enhancement module includes a temporal encoder and a feature enhancement module based on a temporal convolutional network (TCN). The temporal encoder is an improvement on the positional encoder of the Transformer, taking into account the high temporal dependence of EEG signals. Temporal sequence information is also very important to the model, so positional encoding is applied to the time steps to construct a temporal encoder. By marking each time step, the time steps of the mixed features are encoded, thereby enhancing the temporal dependence of the features. The encoding formula of the temporal encoder is as follows:

[0094]

[0095]

[0096] Where ts represents the time step of the data in the sequence, and i refers to the dimension of the data.

[0097] The feature enhancement module based on temporal convolutional network (TCN) consists of five temporal convolutional layers, five pooling layers, and one fully connected layer. Each TCN layer enhances the features with a stride of 16, and finally enhances the features to 128. Finally, a consciousness state index representing the consciousness state is output through a fully connected layer.

[0098] The consciousness monitoring network described above is trained on a dataset using conventional neural network training methods. This dataset uses an internationally standardized anesthesia dataset, primarily containing EEG data and end-tidal concentrations of the anesthetic agent. The EEG data consists of single-channel, one-dimensional data with a sampling frequency of 100Hz. The end-tidal concentration of the anesthetic agent is the concentration of the anesthetic agent exhaled by the body and can reflect the depth of anesthesia to some extent; its sampling frequency is 0.2Hz. While the end-tidal concentration reflects the drug concentration after metabolism, there is a certain deviation between this concentration and the actual depth of anesthesia. To improve the accuracy of anesthesia monitoring, this embodiment employs a more scientific method: based on pharmacokinetic and pharmacodynamic formulas, the effect-site concentration of the drug is calculated using the end-tidal concentration of the anesthetic agent as a label indicating the depth of anesthesia, and this is used for subsequent neural network training. The formula for calculating the effect-site concentration is as follows:

[0099]

[0100] Among them, C et (t) represents the end-tidal sevoflurane concentration at the current moment, which was recorded during the acquisition process at a sampling rate of 0.2 Hz. C eff (t) represents the calculated effect-site concentration, i.e., the label of this anesthesia dataset, k e0 Δt is a constant representing the first-order elimination rate of the effect chamber, and represents the time interval, which is 5s at 0.2Hz.

[0101] Since the sampling frequency of EEG signals is 100Hz, while the sampling frequency of anesthetic end-tidal concentration is 0.2Hz, meaning that 500 EEG data nodes correspond to one labeled data, the EEG data is divided into segments every 5 seconds, corresponding to one labeled data.

[0102] The effectiveness of the proposed method in this embodiment was verified through experiments. The image was input into the consciousness state monitoring network to obtain the consciousness state index. The accuracy of consciousness state monitoring was calculated using labels and compared with different feature groups and other models based on hybrid feature architectures. The comparison results are shown in Table 1.

[0103] Table 1. Experimental Comparison Results of Different Feature Combinations and Hybrid Feature Architecture Models

[0104]

[0105]

[0106] To ensure the rigor of the experiments, all experiments employed the same preprocessing and traditional feature extraction steps, along with the same hybrid feature architecture and time encoder, effectively comparing the prediction performance of different algorithms on the same data. Table 1 shows that wavelet entropy, sample entropy, and differential entropy have the greatest impact on the results. However, permutation entropy and power spectral density (PSD) also have their effects. Combining these two features with wavelet entropy, sample entropy, and differential entropy still improves accuracy, indicating that these two features can supplement wavelet entropy and differential entropy. Therefore, this embodiment ultimately determined to use a combination of entropy features and PSD features as the final feature combination, demonstrating that hybrid features are effective and can improve the accuracy of consciousness state monitoring. Experiments with different models show that the simultaneous use of TCN and hybrid features achieved very high prediction levels when Bi LSTM, Bi GRU, and GCN were used as deep feature extractors. This fully demonstrates the effectiveness of the TCN feature enhancement method designed based on the high time dependence of EEG signals in this embodiment. In conclusion, the consciousness state monitoring method based on GAT and TCN proposed in this embodiment is effective and has good clinical application value.

[0107] Example 2

[0108] One embodiment of this disclosure provides a consciousness state monitoring system based on graph attention and temporal convolution, including a data acquisition module, a graph construction module, and an exponent calculation module:

[0109] The data acquisition module is configured to acquire and segment the EEG data of the person to be monitored.

[0110] The graph construction module is configured to: take each segment of EEG data as a node, extract the entropy features and power spectrum features of each segment of EEG data as node features, and then construct a graph based on the correlation between nodes.

[0111] The index calculation module is configured to input the graph into the trained consciousness state monitoring network to obtain a consciousness state index that represents the consciousness state.

[0112] The consciousness state monitoring network extracts high-dimensional temporal features from the graph through graph attention, fuses the high-dimensional temporal features with node features, enhances the fused hybrid features through temporal convolution, and calculates the consciousness state index using the enhanced hybrid features.

[0113] Example 3

[0114] The purpose of this embodiment is to provide a computer-readable storage medium.

[0115] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the steps in the consciousness state monitoring method based on graph attention and temporal convolution as described in Embodiment 1 of this disclosure.

[0116] Example 4

[0117] The purpose of this embodiment is to provide an electronic device.

[0118] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the consciousness state monitoring method based on graph attention and temporal convolution as described in Embodiment 1 of this disclosure.

[0119] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring conscious states based on graph attention and temporal convolution, characterized in that, include: Acquire and segment the electroencephalogram (EEG) data of the individuals to be monitored; Using each segment of EEG data as a node, the entropy and power spectrum features of each segment of EEG data are extracted as node features, and then a graph is constructed based on the correlation between nodes. The graph is input into the trained consciousness state monitoring network to obtain the consciousness state index, which represents the consciousness state. The consciousness state monitoring network includes a deep feature extraction module, a feature fusion layer, and a feature enhancement module. The deep feature extraction module, based on graph attention, extracts high-dimensional temporal features from the graph. The feature fusion layer fuses the high-dimensional temporal features and node features to obtain hybrid features. The feature enhancement module enhances the hybrid features through time-step-based encoding and temporal convolution-based enhancement. The time-step-based encoding utilizes a temporal encoder to encode the time steps of the hybrid features based on the high temporal dependence of the EEG data. The temporal convolution-based enhancement further enhances the hybrid features using a temporal convolutional network. The enhanced hybrid features are then used to calculate a consciousness state index.

2. The method for monitoring consciousness states based on graph attention and temporal convolution as described in claim 1, characterized in that, The acquisition of the EEG data specifically involves: Perform single-channel EEG data sampling; Segmented according to preset time intervals.

3. The method for monitoring consciousness states based on graph attention and temporal convolution as described in claim 1, characterized in that, It also includes preprocessing the segmented EEG data, including bandpass filtering, baseline drift removal, and notch filtering.

4. The method for monitoring consciousness states based on graph attention and temporal convolution as described in claim 1, characterized in that, The entropy features include four types: wavelet entropy, permutation entropy, sample entropy, and differential entropy; these four entropy features are extracted from each segment of EEG data; The power spectral characteristics are the power spectral densities of the Delta, Theta, Alpha, and Beta bands of the EEG data.

5. The method for monitoring consciousness states based on graph attention and temporal convolution as described in claim 4, characterized in that, The construction graph is specifically as follows: The correlation between each EEG data segment is calculated as the adjacency matrix, and the entropy feature and power spectrum feature of each EEG data segment are used as node features to construct a node feature matrix, resulting in a graph composed of the adjacency matrix and the node feature matrix.

6. A consciousness state monitoring system based on graph attention and temporal convolution, characterized in that, It includes a data acquisition module, a graph construction module, and an index calculation module: The data acquisition module is configured to acquire and segment the EEG data of the person to be monitored. The graph construction module is configured to: take each segment of EEG data as a node, extract the entropy features and power spectrum features of each segment of EEG data as node features, and then construct a graph based on the correlation between nodes. The index calculation module is configured to input the graph into the trained consciousness state monitoring network to obtain a consciousness state index that represents the consciousness state. The consciousness state monitoring network includes a deep feature extraction module, a feature fusion layer, and a feature enhancement module. The deep feature extraction module, based on graph attention, extracts high-dimensional temporal features from the graph. The feature fusion layer fuses the high-dimensional temporal features and node features to obtain hybrid features. The feature enhancement module enhances the hybrid features through time-step-based encoding and temporal convolution-based enhancement. The time-step-based encoding utilizes a temporal encoder to encode the time steps of the hybrid features based on the high temporal dependence of the EEG data. The temporal convolution-based enhancement further enhances the hybrid features using a temporal convolutional network. The enhanced hybrid features are then used to calculate a consciousness state index.

7. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. Processor, for executing the computer-readable instructions; When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-5.

8. A storage medium, characterized in that, The computer-readable instructions are stored non-temporarily, wherein when the computer-readable instructions are executed by a computer, the method described in any one of claims 1-5 is performed.

Images