Seizure detection method, apparatus and network device based on feature fusion

Abstract

The embodiment of the application provides a feature fusion-based seizure detection method and device and network equipment, the method comprises the following steps: acquiring channel data of each channel of electroencephalogram signal, and performing feature extraction on the channel data according to a convolutional neural network to obtain first features of the channel; performing correlation analysis between channels according to the first features to determine second features between the channels; fusing the first features and the second features to obtain fused features; inputting the fused features into a classification model containing a multi-head attention mechanism to determine a seizure analysis result. The method fuses the features in the channel and the features between the channels, comprehensively considers the channel data itself and the correlation between the channel data, and can more accurately identify the seizure.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer information, in particular, the present application relates to a kind of seizure detection method, device and network equipment based on feature fusion. BACKGROUND

[0002] Epilepsy is a serious nervous system disease. It is a time-consuming and laborious task to detect seizures from long-term electroencephalogram monitoring. In addition, different types of seizures usually originate from different brain regions, and seizures show complexity and diversity. These factors make accurate seizure detection a key but challenging task in clinical practice. Electroencephalogram (EEG) signals, as an important tool for recording brain activity, are widely used in seizure detection.

[0003] Existing solutions only consider intracanal features of electroencephalogram, without considering interchannel features of electroencephalogram, resulting in low accuracy of recognition results. SUMMARY

[0004] Embodiments of the present application provide a seizure detection method, device and network equipment based on feature fusion, which can analyze intracanal features of electroencephalogram and interchannel features of electroencephalogram, and analyze after fusion, to more accurately identify seizures. The technical solution is as follows:

[0005] In a first aspect, the present application provides a seizure detection method based on feature fusion, the method comprising: obtaining channel data of each electroencephalogram signal channel, and extracting features of the channel data according to a convolutional neural network to obtain first features of the channel; performing interchannel correlation analysis according to the first features to determine second features between the channels; fusing the first features and the second features to obtain fused features; inputting the fused features into a classification model containing a multi-head attention mechanism to determine a seizure analysis result.

[0006] Further, the interchannel correlation analysis according to the first features to determine the second features between the channels comprises: analyzing the Pearson correlation coefficient between the channels according to the first features to perform interchannel correlation analysis and obtain a feature matrix between the channels; incorporating adaptive adjustable weight information into the feature matrix, and learning the weight information according to the first features to determine the second features between the channels, the weight information corresponding to the correlation between the channels.

[0007] Further, the first feature is analyzed according to the Pearson correlation coefficient between the channels to perform inter-channel correlation analysis to obtain a feature matrix between the channels, including: obtaining first data of the first feature of the first channel and second data of the second feature of the second channel; obtaining a first average value of the first channel and a second average value of the second channel; analyzing the Pearson correlation coefficient between the channels according to the first data, the second data, the first average value, and the second average value to form a correlation relationship matrix between the channels; and processing the values in the correlation relationship matrix according to a preset threshold to obtain the feature matrix between the channels; and after the feature matrix is processed by the preset threshold, the values greater than or equal to the preset threshold in the correlation relationship matrix are 1, and the values less than the preset threshold are 0.

[0008] Further, the weight information that can be adaptively adjusted is integrated into the feature matrix, and learning of the weight information is performed according to the first feature to determine the second feature between the channels, including: generating a diagonal matrix of learnable parameters, the parameters in the diagonal matrix being related to the weight information of the correlation relationship between the channels; integrating the diagonal matrix into the feature matrix to obtain a self-loop perception matrix, and performing learning according to the first feature; calculating a degree matrix of the self-loop perception matrix, and performing normalization processing according to the self-loop perception matrix and the degree matrix to obtain a symmetric normalized Laplacian matrix as the second feature between the channels.

[0009] Further, the first feature and the second feature are fused to obtain a fusion feature, including: performing hierarchical linear transformation processing on the first feature and the second feature to obtain an intermediate representation; and performing activation processing on the intermediate representation by using a hyperbolic tangent function to obtain the fusion feature.

[0010] Further, the hierarchical linear transformation processing on the first feature and the second feature to obtain the intermediate representation includes: performing hierarchical linear transformation processing on the first feature and the second feature by using a first fully connected layer to obtain an intermediate matrix; and performing processing on the intermediate matrix by using a second fully connected layer to obtain the intermediate representation.

[0011] Further, the classification model comprising the multi-head attention mechanism is configured to: obtain time information of the fusion feature as a time feature; determine relevant features of the fusion feature according to the time feature of the fusion feature; determine attention weights corresponding to the relevant features before and after the fusion feature, and integrate the relevant features into the fusion feature according to the attention weights to determine a pooling feature; and perform processing on the pooling feature by using an activation function to determine an epilepsy analysis result.

[0012] Further, the channel data of each channel of the brain electrical signals comprises: acquiring channel related information of each channel of the brain electrical signals, and performing band-pass filtering in a preset frequency band; the channel related information comprises electroencephalogram data and time information, and the channel related information comprises first information corresponding to a seizure and second information corresponding to a non-seizure; the filtered channel related information is segmented according to the time information and a preset time interval to obtain information segments, and normalized processing is performed to obtain the channel data.

[0013] In a second aspect, the application provides a seizure detection device based on feature fusion, comprising: a first feature acquisition module, configured to acquire channel data of each channel of the brain electrical signals, and perform feature extraction on the channel data according to a convolutional neural network to obtain first features of the channels; a second feature acquisition module, configured to perform correlation analysis between the channels according to the first features to determine second features between the channels; a fusion feature acquisition module, configured to fuse the first features and the second features to obtain fusion features; and a classification model analysis module, configured to input the fusion features into a classification model containing a multi-head attention mechanism to determine a seizure analysis result.

[0014] In a third aspect, the application provides a network device, comprising: a memory, a transceiver, and a processor; wherein the memory is configured to store a computer program; the transceiver is configured to transceive data under the control of the processor; and the processor is configured to read the computer program in the memory and execute the method according to the first aspect.

[0015] In a fourth aspect, the application provides a storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method according to the first aspect.

[0016] The technical scheme provided by the application has the beneficial effects that:

[0017] The proposed solution can be applied to scenarios involving EEG data-based disease identification. For example, it can be used to identify whether a patient is experiencing epilepsy, allowing for timely alerts and interventions. This solution acquires channel data from various EEG signal channels, extracts channel features, analyzes inter-channel correlations to extract inter-channel features, and then fuses these features with a classification model to analyze the fused characteristics and determine if an epileptic seizure has occurred. By fusing intra-channel and inter-channel features, this solution comprehensively considers both the channel data itself and the correlations between them, enabling more accurate identification of epileptic seizures. Specifically, this solution acquires channel data from various EEG signal channels, extracts features from the channel data using a convolutional neural network to obtain the first channel feature, performs inter-channel correlation analysis based on the first feature to determine the second inter-channel feature, fuses the first and second features to obtain the fused feature, and inputs the fused feature into a classification model incorporating multi-head attention mechanisms to analyze the correlation between the features before and after fusion and determine the epilepsy analysis result. This approach pre-trains convolutional neural networks and classification models incorporating multi-head attention mechanisms for feature extraction and analysis. It considers both the spatial correlation between channels and the temporal correlation of features across different channels, enabling more accurate epilepsy analysis. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0019] Figure 1 This is a schematic flowchart of a feature fusion-based epileptic seizure detection method according to an embodiment of this application;

[0020] Figure 2 This is a schematic flowchart of another embodiment of the epileptic seizure detection method based on feature fusion in this application;

[0021] Figure 3 This is a schematic diagram of the structure of a feature fusion-based epileptic seizure detection device according to an embodiment of this application;

[0022] Figure 4 This is a structural block diagram of a network device according to an embodiment of this application;

[0023] Figure 5 This is a structural block diagram of a user equipment according to one embodiment of this application. Detailed Implementation

[0024] Embodiments of the present application are described below in detail with reference to the accompanying drawings, examples of which are shown in the drawings, wherein the same or similar notations identify the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0025] Those skilled in the art can understand that, unless specifically stated, the singular forms "a", "an", "said" and "the" used herein also include the plural forms, and "multiple" refers to two or more, and other quantifiers similar thereto. It should be further understood that the phrase "comprising" used in the specification of the present application means that the features, integers, steps, operations, elements and / or components exist, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we say that an element is "connected" or "coupled" to another element, it can be directly connected or coupled to the other element, or there can be intermediate elements. In addition, "connected" or "coupled" used herein can include wireless connection or wireless coupling. The phrase "and / or" used herein describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. The character " / " generally represents that the associated objects before and after are in an "or" relationship.

[0026] The present application proposes a seizure detection method, which first uses the Pearson correlation coefficient combined with a learnable self-loop weight to dynamically adjust the channel weight in the training process to extract inter-channel features, and then combines these features with intra-channel features to obtain a more comprehensive view of brain activity. Then, a graph convolution network based on self-attention mechanism (LATGCN) is proposed to identify inter-channel and intra-channel features of electroencephalogram channels to effectively capture spatial and temporal information for seizure detection to assist doctors in the process of clinical diagnosis of epilepsy to achieve seizure detection.

[0027] Specifically, in order to improve the accuracy of seizure detection, the present application proposes a seizure detection method, the flow chart of which is shown in Figure 1 The whole process is divided into four stages: data preprocessing, inter-channel feature extraction, inter-channel and intra-channel feature fusion, and seizure detection.

[0028] Data preprocessing: The electroencephalogram (EEG) data is band-pass filtered between 4-30 Hz, which covers important frequency components closely related to brain activity, such as theta waves (4-7 Hz), alpha waves (8-12 Hz), and beta waves (13-30 Hz). Then, the electroencephalogram recordings of seizures and non-seizures are segmented into 2-second intervals to reduce the computational burden of the model. Finally, the EEG segments are normalized to eliminate amplitude differences between different recordings.

[0029] Inter-channel and intra-channel feature fusion: In electroencephalogram analysis, inter-channel features highlight the functional connections and interactions between different brain regions, revealing how these brain regions coordinate during different cognitive processes. Analysis of inter-channel features helps detect seizures from long-term electroencephalogram signal monitoring. The Pearson correlation coefficient is a commonly used measure of functional connectivity that quantifies the strength and direction of linear relationships between two signals, revealing functional coupling between different brain regions. Therefore, the present application extracts inter-channel features by calculating the Pearson correlation coefficient of different electroencephalogram signal channels.

[0030] First, the preprocessed electroencephalogram signals are input into a convolutional neural network to extract features from each channel. Subsequently, the Pearson correlation coefficient between each channel is calculated after collecting all features of each channel. The convolutional neural network here consists of three convolutional layers, three pooling layers, and one fully connected layer for extracting channel features. To enable the network to accurately extract channel features, the network is pre-trained until the training loss converges and the validation accuracy exceeds 90%. The Pearson correlation coefficient is calculated as follows:

[0031]

[0032] In formula 1, x i and y i are the data values corresponding to the two channels, and are the average values of the data of the two channels. The Pearson correlation coefficient has a value range of [-1, 1], a value close to 1 indicates a strong positive correlation between the two variables, a value close to -1 indicates a strong negative correlation, and a value of 0 indicates no linear correlation between the variables. To extract as many features as possible from brain regions with strong functional connections and filter out interfering brain regions with weak connections, the present application sets a threshold for the correlation matrix, values above the threshold are set to 1, indicating that there is a direct connection between the channels, values below the threshold are set to 0, indicating that there is no connection between the channels, and finally the inter-channel feature matrix can be obtained.

[0033] Secondly, based on the extracted features, an inter-channel feature matrix is calculated, and a learnable Laplacian matrix with self-loop is introduced, called self-aware Laplacian matrix, so that the network can adaptively adjust these weights during the training process. For a given N x N inter-channel feature matrix A, add it to the diagonal matrix S containing learnable parameters θ s = [θ s1 , θ s2 ..., θ sN ] to obtain the self-aware Laplacian matrix And calculate the degree matrix The calculation formula is as follows:

[0034]

[0035] The results calculated by the above formula 2 and formula 3 can be used to further calculate the symmetric normalized Laplacian matrix L, and the calculation formula is as follows:

[0036]

[0037] In the formula, I represents the unit matrix.

[0038] The symmetric normalized Laplacian matrix is used to process the input features, and the inter-channel information is integrated through matrix multiplication to form an intermediate representation. Then, linear transformation is performed through two fully connected layers, and activated by the hyperbolic tangent function. This layered linear transformation can further mine and integrate the node features in the graph, improving the convergence speed of the model. The node feature update formula of each layer of convolutional neural network (GCN) is as follows:

[0039]

[0040] In the formula, H i represents the i-th layer node feature matrix of GCN, and are the weight matrices of the two fully connected layers of the i-th layer, and H represents the intermediate features after the first linear transformation.

[0041] Finally, through the Laplacian matrix multiplication and layered linear transformation, the fusion features of intra-channel and inter-channel are obtained, so that the overall view of brain activity covering local dynamics and global network characteristics can be obtained.

[0042] Seizure detection based on self-aware Transformer-graph convolutional network: in the self-aware Transformer rIn the graph convolution network (LATGCN), firstly, GCN layers are used to process the relationship between nodes, and the fusion features within and between channels are used as the input of the network. Secondly, the Transformer encoder after the GCN layer can obtain time information. The Transformer encoder contains an attention mechanism, which enables each input token to interact with other tokens, thereby effectively capturing global temporal dependencies. The multi-head attention mechanism allows the model to simultaneously focus on different temporal features of the input through queries, keys, and values. The position feedforward network independently processes each time position, further improving the expression ability of the time features. The residual connection and layer normalization enhance the stability of the model, and the position encoding provides the model with position information of the time sequence. These components work together to enable the Transformer encoder to effectively model complex temporal dependencies.

[0043] The features of all nodes are aggregated by a pooling layer with an attention mechanism, and the pooled feature vector p is activated by a Sigmoid function after passing through a fully connected layer to obtain the final prediction result. The calculation formula of the attention pooling layer is as follows:

[0044] p = ∑ softmax(a(x i ))·x i Equation 7

[0045] In the formula, a(x i ) is the attention weight assigned by the attention mechanism to each node feature x i . The attention mechanism can capture features conducive to electroencephalogram recognition and filter out irrelevant features, which can improve the recognition accuracy of electroencephalograms and achieve precise detection of seizures.

[0046] Specifically, the present application provides a seizure detection method based on feature fusion, as shown in Figure 2 , the method comprises:

[0047] Step 202, acquiring channel data of channels of each electroencephalogram signal, and performing feature extraction on the channel data according to a convolutional neural network to obtain first features of the channels. The first features can also be referred to as channel features or intra-channel features.

[0048] Step 204, performing correlation analysis between channels based on the first features to determine second features between the channels. The second features can also be referred to as inter-channel features.

[0049] Step 206, fusing the first features and the second features to obtain fusion features.

[0050] Step 208, inputting the fusion features into a classification model containing a multi-head attention mechanism to determine a seizure analysis result.

[0051] It should be noted that part of the implementation process of the scheme has been described in the above embodiment, and the implementation process of the above embodiment can be referred to, which will not be repeated here.

[0052] The scheme of the present application can be applied to the scene of identifying a disease based on electroencephalogram data. For example, whether a patient has a seizure can be identified based on electroencephalogram data, so as to timely give an alarm and the like. The scheme can obtain channel data of channels of electroencephalogram signals, extract channel features, analyze the correlation between channels, extract inter-channel features, and then fuse the channel features and the inter-channel features, so as to analyze the fused features by using a classification model to determine whether a seizure occurs. The scheme fuses the intra-channel features and the inter-channel features, comprehensively considers the channel data itself and the correlation between the channel data, and can more accurately identify the occurrence of a seizure. Specifically, the scheme can obtain channel data of channels of electroencephalogram signals, extract channel features from the channel data by using a convolutional neural network to obtain first features of the channels, analyze the correlation between the channels based on the first features to determine second features between the channels, fuse the first features and the second features to obtain fused features, and input the fused features into a classification model containing a multi-head attention mechanism to analyze the correlation features before and after the fused features and determine a seizure analysis result. The scheme can pre-train the convolutional neural network and the classification model containing the multi-head attention mechanism to extract features and analyze features.

[0053] The scheme can extract channel features corresponding to channel data by using a convolutional neural network, analyze Pearson correlation coefficients between channels, and fuse learnable weight information, and then analyze the correlation between the channels as inter-channel features. Specifically, as an optional embodiment, the analysis of the correlation between the channels based on the first features to determine the second features between the channels includes: analyzing Pearson correlation coefficients between the channels based on the first features to analyze the correlation between the channels and obtain a feature matrix between the channels; and fusing self-adaptive weight information in the feature matrix and learning the weight information based on the first features to determine the second features between the channels, the weight information corresponding to the correlation between the channels.

[0054] The scheme can determine the correlation matrix by analyzing the channel data and the channel average value, and convert the values greater than or equal to the preset threshold in the correlation matrix to 1 and the values less than the preset threshold to 0 according to the preset threshold to form the feature matrix. Specifically, as an optional embodiment, the Pearson correlation coefficient between the channels is analyzed according to the first feature to analyze the correlation between the channels, and the feature matrix between the channels is obtained, including: obtaining first data of the first feature of the first channel and second data of the second feature of the second channel; obtaining the first average value of the first channel and the second average value of the second channel; analyzing the Pearson correlation coefficient between the channels according to the first data, the second data, the first average value and the second average value to form the correlation matrix between the channels; processing the values in the correlation matrix according to the preset threshold to obtain the feature matrix between the channels; after the feature matrix is processed by the preset threshold, the values greater than or equal to the preset threshold in the correlation matrix are 1, and the values less than the preset threshold are 0.

[0055] After obtaining the feature matrix, a diagonal matrix of learnable parameters is added to the feature matrix, and learning is performed according to the training data to determine the parameters corresponding to the diagonal matrix to determine the second feature. Specifically, as an optional embodiment, the weight information that can be adaptively adjusted is integrated into the feature matrix, and the learning of the weight information is performed according to the first feature to determine the second feature between the channels, including: generating a diagonal matrix of learnable parameters, the parameters in the diagonal matrix being related to the weight information of the correlation between the channels; integrating the diagonal matrix into the feature matrix to obtain a self-loop perception matrix, and performing learning according to the first feature; calculating the degree matrix of the self-loop perception matrix, and performing normalization processing according to the self-loop perception matrix and the degree matrix to obtain a symmetric normalized Laplacian matrix as the second feature between the channels.

[0056] The scheme can perform linear transformation through two fully connected layers and adopt hyperbolic tangent function activation to form the fusion feature. Specifically, as an optional embodiment, the first feature and the second feature are fused to obtain the fusion feature, including: performing hierarchical linear transformation processing on the first feature and the second feature to obtain intermediate representation; performing activation processing on the intermediate representation by adopting hyperbolic tangent function to obtain the fusion feature. Specifically, as an optional embodiment, the hierarchical linear transformation processing is performed on the first feature and the second feature to obtain the intermediate representation, including: performing hierarchical linear transformation processing on the first feature and the second feature through the first fully connected layer to obtain an intermediate matrix; performing processing on the intermediate matrix through the second fully connected layer to obtain the intermediate representation.

[0057] After learning the spatial features between the channels, the scheme can also learn the association of the fusion features in time according to a pre-trained classification model containing a multi-head attention mechanism, and integrate the features before and after the fusion features to form a pooling feature for analysis. Specifically, as an optional embodiment, the classification model containing the multi-head attention mechanism is configured to: obtain time information of the fusion features as time features; determine relevant features of the fusion features according to the time features of the fusion features; determine attention weights of the relevant features before and after the fusion features, and integrate the relevant features into the fusion features according to the attention weights to determine a pooling feature; and process the pooling feature by using an activation function to determine an epilepsy analysis result.

[0058] Before analyzing the data, the scheme can be preprocessed. Specifically, as an optional embodiment, the channel data of each channel of the brain electrical signals includes: obtaining channel-related information of each channel of the brain electrical signals, and performing band-pass filtering in a preset frequency band; the channel-related information includes electroencephalogram data and time information, and the channel-related information includes first information corresponding to a seizure and second information corresponding to a non-seizure; the filtered channel-related information is segmented according to the time information and a preset time interval to obtain information segments, and normalized to obtain channel data. The preset frequency band is 4-30 Hz, and the preset time interval can be 2s.

[0059] Based on the above embodiments, the present application also provides a seizure detection device based on feature fusion, as shown in Figure 3 The device includes:

[0060] The first feature acquisition module 302 is configured to obtain channel data of each channel of the brain electrical signals, and extract features of the channels by using a convolutional neural network to obtain first features of the channels.

[0061] The second feature acquisition module 304 is configured to analyze the correlation between the channels according to the first features to determine second features between the channels.

[0062] The fusion feature acquisition module 306 is configured to fuse the first features and the second features to obtain fusion features.

[0063] The classification model analysis module 308 is configured to input the fusion features into a classification model containing a multi-head attention mechanism to determine an epilepsy analysis result.

[0064] The implementation manner of the embodiments of the present application is similar to that of the above method embodiments, and the specific implementation manner can refer to the specific implementation manner of the above method embodiments, which will not be repeated here.

[0065] The scheme of the present application can be applied to the scene of identifying a disease based on electroencephalogram data. For example, whether a patient has a seizure can be identified based on electroencephalogram data, so as to timely give an alarm and the like. The scheme can obtain channel data of channels of electroencephalogram signals, extract channel features, analyze the correlation between channels, extract inter-channel features, and then fuse the channel features and the inter-channel features, so as to analyze the fused features by using a classification model to determine whether a seizure occurs. The scheme fuses the intra-channel features and the inter-channel features, comprehensively considers the channel data itself and the correlation between the channel data, and can more accurately identify the occurrence of a seizure. Specifically, the scheme can obtain channel data of channels of electroencephalogram signals, extract features of the channels by using a convolutional neural network, obtain first features of the channels, analyze the correlation between the channels based on the first features to determine second features between the channels, fuse the first features and the second features to obtain fused features, and input the fused features into a classification model containing a multi-head attention mechanism to analyze the relevant features before and after the fused features and determine a seizure analysis result. The scheme can pre-train the convolutional neural network and the classification model containing the multi-head attention mechanism to perform feature extraction and feature analysis.

[0066] It should be noted that the division of units and / or modules in the embodiments of the present application is illustrative, and is only a logical function division. When actually implemented, another division mode can be used. In addition, each functional unit and / or module in each embodiment of the present application can be integrated in one processing unit and / or module, or each unit and / or module can be physically present alone, or two or more units and / or modules can be integrated in one unit and / or module. The integrated unit and / or module can be realized in the form of hardware or in the form of a software functional unit and / or module.

[0067] When the integrated unit and / or module is realized in the form of a software functional unit and / or module and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical scheme of the present application or the part that essentially contributes to the related art or the whole or part of the technical scheme can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0068] In addition, the data transmission device and the data transmission method provided by the above embodiments are based on the same application concept, and the implementation of the device and the method can be referred to each other because the principles of solving problems are similar, and the repeated parts will not be described again.

[0069] Figure 4 A structural block diagram of a network device according to an exemplary embodiment is shown.

[0070] As shown in Figure 4 , the network device 1100 at least includes a processor 1110, a memory 1120 and a transceiver 1130.

[0071] The transceiver 1130 is configured to receive and send data under the control of the processor 1110.

[0072] In Figure 4 , the bus architecture can include any number of interconnected buses and bridges, which link various circuits together, including one or more processors represented by the processor 1110 and the memory represented by the memory 1120. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, and thus, will not be further described herein. The bus interface provides an interface. The transceiver 1130 can be a plurality of elements, i.e., including a transmitter and a receiver, which provide units and / or modules for communicating with various other devices on transmission media, including wireless channels, wired channels, optical cables, etc.

[0073] The processor 1110 is responsible for managing the bus architecture and general processing, and the memory 1120 can store data used by the processor 1110 when performing operations.

[0074] Optionally, the processor 1110 can be a central processor (CPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or a complex programmable logic device (CPLD), and the processor 1110 can also adopt a multi-core architecture. The processor 1110 and the memory 1120 can also be arranged physically separately.

[0075] The processor 1110, by invoking the computer program stored in the memory 1120, is configured to execute the executable instructions obtained to perform any one of the allocation methods of the cell radio network temporary identifier provided by the above embodiments.

[0076] Figure 5 A structural block diagram of a user equipment according to an exemplary embodiment is shown.

[0077] As shown in Figure 5 , the user equipment 1300 at least includes a processor 1310, a memory 1320 and a transceiver 1330.

[0078] The transceiver 1330 is configured to receive and send data under the control of the processor 1310.

[0079] In Figure 5 , the bus architecture can include any number of interconnected buses and bridges, which link together various circuits, including one or more processors, represented by the processor 1310, and the memory, represented by the memory 1320. The bus architecture can also link together various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and thus, are not further described herein. The bus interface provides an interface. The transceiver 1330 can be a plurality of elements, i.e., including a transmitter and a receiver, which provide units and / or modules for communicating with various other apparatuses over transmission media, including wireless channels, wired channels, optical cables, etc. The user interface 1340 can also be an interface capable of connecting to external or internal devices, including but not limited to a keypad, a display, a speaker, a microphone, a joystick, etc., for different user equipment.

[0080] The processor 1310 is responsible for managing the bus architecture and general processing, and the memory 1320 can store data used by the processor 1310 in performing operations.

[0081] Optionally, the processor 1310 can be a CPU (Central Processor Unit), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or a CPLD (Complex Programmable Logic Device), and the processor 1310 can also adopt a multi-core architecture. The processor 1310 and the memory 1320 can also be physically arranged separately.

[0082] The processor 1310, by invoking a computer program stored in the memory 1320, is configured to execute the executable instructions obtained to perform any one of the allocation methods of the cell radio network temporary identifier provided by the above embodiments.

[0083] It should be noted that the above device provided by the embodiments of the present application can realize all the method steps achieved by the above method embodiments, and achieve the same technical effects. Therefore, the same parts and beneficial effects of the method embodiments will not be described in detail.

[0084] In addition, a storage medium is provided in the embodiments of the present application, and the storage medium stores a computer program. The computer program is executed by a processor to implement the data transmission method in the above embodiments. The storage medium can be any available medium or data storage device accessible by the processor, including but not limited to a magnetic storage (such as a floppy disk, a hard disk, a magnetic tape, a magneto-optical disk (MO), etc.), an optical storage (such as a CD, a DVD, a BD, a HVD, etc.), and a semiconductor memory (such as a ROM, an EPROM, an EEPROM, a non-volatile memory (NAND FLASH), a solid-state disk (SSD), etc.).

[0085] In the embodiments of the present application, a program product is provided, for example, the program product is an FPGA chip or a DSP chip. The program product includes executable instructions stored in a storage medium. The processor reads the executable instructions from the storage medium, so that the executable instructions are executed by the processor to implement the data transmission method in the above embodiments.

[0086] The scheme of the present application can be applied to the scene of identifying a disease based on electroencephalogram data. For example, whether a patient has a seizure can be identified based on electroencephalogram data, so as to timely give an alarm and the like. The scheme can obtain channel data of channels of electroencephalogram signals, extract channel features, analyze the correlation between channels, extract inter-channel features, and then fuse the channel features and the inter-channel features, so as to analyze the fused features by using a classification model to determine whether a seizure occurs. The scheme fuses the intra-channel features and the inter-channel features, comprehensively considers the channel data itself and the correlation between the channel data, and can more accurately identify the occurrence of a seizure. Specifically, the scheme can obtain channel data of channels of electroencephalogram signals, extract features from the channel data based on a convolutional neural network to obtain first features of the channels, analyze the correlation between the channels based on the first features to determine second features between the channels, fuse the first features and the second features to obtain fused features, and input the fused features into a classification model containing a multi-head attention mechanism to analyze the correlation features before and after the fused features and determine a seizure analysis result. The scheme can pre-train the convolutional neural network and the classification model containing the multi-head attention mechanism to perform feature extraction and feature analysis.

[0087] Those skilled in the art will appreciate that embodiments of the application can be readily used as a method, a system or a computer program product. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, magnetic disks and optical storage media) embodying computer readable program code.

[0088] The application is described in reference to the flowchart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer executable instructions. The computer executable instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart Figure 1 one or more functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks.

[0089] These computer executable instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks.

[0090] These computer executable instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks. Figure 1 one or more functions specified in the flowchart block or blocks.

[0091] It should be understood that although the steps in the flowcharts of the drawings are shown in a sequential order following the arrows, the steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated otherwise herein, the execution of the steps is not strictly limited to the order indicated by the arrows, and can be executed in other orders. Moreover, at least some of the steps in the flowcharts of the drawings can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of which is not necessarily sequential, but can be round-robin or alternating with at least some of the other steps or sub-steps or stages of other steps.

[0092] The above only describes some embodiments of the present application, and it should be pointed out that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application.

Claims

1. A seizure detection method based on feature fusion, characterized in that, The method comprises: obtaining channel data of each brain electrical signal channel, and performing feature extraction on the channel data according to a convolutional neural network to obtain first features of the channels; performing correlation analysis between the channels according to the first features to determine second features between the channels; fusing the first features and the second features to obtain fused features; inputting the fused features into a classification model comprising a multi-head attention mechanism to determine an epilepsy analysis result; the correlation analysis between the channels according to the first features to determine the second features between the channels comprises: analyzing the Pearson correlation coefficients between the channels according to the first features to perform the correlation analysis between the channels and obtain a feature matrix between the channels; fusing adaptive weight information into the feature matrix, and learning the weight information according to the first features to determine the second features between the channels, the weight information corresponding to the correlation between the channels; the fusing of the adaptive weight information into the feature matrix and the learning of the weight information according to the first features to determine the second features between the channels comprises: generating a diagonal matrix of learnable parameters, the parameters in the diagonal matrix being related to the weight information of the correlation between the channels; fusing the diagonal matrix into the feature matrix to obtain a self-loop awareness matrix, and learning according to the first features; calculating a degree matrix of the self-loop awareness matrix, and performing normalization processing on the self-loop awareness matrix and the degree matrix to obtain a symmetric normalized Laplacian matrix as the second features between the channels the classification model comprising the multi-head attention mechanism is used to: obtain time information of the fused features as time features; determine relevant features of the fused features according to the time features of the fused features; determine attention weights corresponding to the relevant features before and after the fused features, and fuse the relevant features into the fused features according to the attention weights to determine a pooling feature; determine the epilepsy analysis result by processing the pooling feature with an activation function.

2. The method of claim 1, wherein, the analysis of the Pearson correlation coefficients between the channels according to the first features to perform the correlation analysis between the channels and obtain the feature matrix between the channels comprises: obtaining first data of the first features of the first channel and second data of the second features of the second channel; obtaining a first average value of the first channel and a second average value of the second channel; analyzing the Pearson correlation coefficients between the channels according to the first data, the second data, the first average value, and the second average value to form a correlation relationship matrix between the channels; processing the values in the correlation relationship matrix according to a preset threshold to obtain the feature matrix between the channels; after the processing of the preset threshold, the values greater than or equal to the preset threshold in the correlation relationship matrix are 1, and the values less than the preset threshold are 0.

3. The method of claim 2, wherein, the fusing of the first features and the second features to obtain the fused features comprises: performing hierarchical linear transformation processing on the first features and the second features to obtain an intermediate representation; performing activation processing on the intermediate representation with a hyperbolic tangent function to obtain the fused features.

4. The method of claim 3, wherein, the hierarchical linear transformation processing on the first features and the second features to obtain the intermediate representation comprises: The first feature and the second feature are processed by a first fully connected layer to obtain an intermediate matrix; The intermediate matrix is processed by a second fully connected layer to obtain an intermediate representation.

5. The method of claim 1, wherein, The channel data of each channel of the electroencephalogram signal is obtained, including: The channel correlation information of each channel of the electroencephalogram signal is obtained, and band-pass filtering is performed in a preset frequency band; the channel correlation information includes electroencephalogram data and time information, and the channel correlation information includes first information corresponding to a seizure and second information corresponding to a non-seizure; The filtered channel correlation information is segmented according to the time information and a preset time interval to obtain information segments, and normalized to obtain channel data.

6. A seizure detection apparatus based on feature fusion, characterized in that, The device comprises: A first feature acquisition module is configured to acquire channel data of each channel of the electroencephalogram signal, and extract features of the channel data based on a convolutional neural network to obtain first features of the channel; A second feature acquisition module is configured to analyze the correlation between channels based on the first features to determine second features between the channels; A fusion feature acquisition module is configured to fuse the first features and the second features to obtain fusion features; A classification model analysis module is configured to input the fusion features into a classification model comprising a multi-head attention mechanism to determine a seizure analysis result; The correlation between channels is analyzed based on the first features to determine the second features between the channels, including: The Pearson correlation coefficient between channels is analyzed based on the first features to analyze the correlation between channels and obtain a feature matrix between the channels; The weight information is learned based on the first features to determine the second features between the channels, and the weight information corresponds to the correlation between the channels; The weight information is learned based on the first features to determine the second features between the channels, including: A diagonal matrix of learnable parameters is generated, and the parameters in the diagonal matrix are related to the weight information of the correlation between the channels; The diagonal matrix is fused into the feature matrix to obtain a self-loop perception matrix, and learning is performed based on the first features; The degree matrix of the self-loop perception matrix is calculated, and the self-loop perception matrix and the degree matrix are normalized to obtain a symmetric normalized Laplacian matrix as the second features between the channels The classification model comprising a multi-head attention mechanism is configured to: Obtain the time information of the fusion features as time features; Determine the related features of the fusion features based on the time features of the fusion features; Determine the attention weights of the related features before and after the fusion features, and fuse the related features into the fusion features according to the attention weights to determine a pooling feature; An activation function is used to process the pooling feature to determine a seizure analysis result.

7. A network device, comprising: It comprises: A memory, a transceiver, and a processor; wherein the memory is used to store a computer program; the transceiver is used to transceive data under the control of the processor; The processor is configured to read the computer program in the memory and execute the method of any one of claims 1-5.

Images