A medical diagnosis assistance system based on EEG signals and artificial intelligence classification

By constructing an appropriate mental state paradigm and a convolutional bidirectional long and short-term memory neural network using wavelet change decomposition and multi-headed attention mechanism, the accuracy and feature fusion of EEG signal classification in psychiatric medical diagnostic auxiliary system is solved, and the problem of improvement of the fusion of EEG signal classification is improved The system's automation and diagnostic auxiliary effect.

CN117612710BActive Publication Date: 2024-12-03HEFEI FOURTH PEOPLES HOSPITAL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410077209.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-19
Publication Date
2024-12-03
Estimated Expiration
2044-01-19

AI Technical Summary

Technical Problem

There is a suitable paradigm for psychiatric diagnosis in the existing psychiatric medical diagnosis system. The characteristics of the subjects are complex and diverse, which affect the accuracy of classification and prediction tasks, and it is difficult to perform complex EEG feature information. Simple feature fusion affects high-precision classification.

Method used

Construct the audio-visual mental state paradigm and the resting mental state paradigm, use discrete wavelet changes to decompose the EEG signal and calculate the signal characteristic information, combine the convolutional bidirectional long and short-term memory neural network with multi-head attention mechanism to classify the EEG signal to realize the characteristic information multimodal fusion and high-precision classification.

Benefits of technology

It improves the accuracy of the classification of artificial intelligence of EEG signals and the automation of the system, provides better theoretical support and data technology, and provides assistance for psychiatric medical diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117612710B_ABST
    Figure CN117612710B_ABST
Patent Text Reader

Abstract

The present invention discloses a medical diagnosis assistance system based on EEG signals and artificial intelligence classification, including an EEG signal acquisition module, a mental state classification paradigm construction module, an EEG information preprocessing module, a feature information processing module, an EEG signal classification module and a medical diagnosis assistance module. The present invention belongs to the technical field of psychiatric medical diagnosis assistance systems, specifically a medical diagnosis assistance system based on EEG signals and artificial intelligence classification. The present invention constructs an audio-visual mental state paradigm and a resting mental state paradigm, which provides good theoretical support for EEG signal acquisition; uses discrete wavelet changes to decompose EEG signals and calculate signal feature information, which improves the overall accuracy of EEG signal artificial intelligence classification; uses a convolutional bidirectional long short-term memory neural network method combined with a multi-head attention mechanism to classify EEG signals, which provides data and technical support for psychiatric medical diagnosis.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of psychiatric medical diagnosis auxiliary system, and specifically is a medical diagnosis auxiliary system based on electroencephalogram signal and artificial intelligence classification. Background Art

[0002] The psychiatric medical diagnosis assistance system based on EEG signals and artificial intelligence classification refers to a system that uses EEG data and artificial intelligence classification algorithms to assist psychiatrists in diagnosing and treating mental illnesses. It combines the biological characteristics of EEG signals with the analysis capabilities of artificial intelligence, aiming to improve the diagnostic accuracy and treatment effectiveness of psychiatrists.

[0003] However, in the existing psychiatric medical diagnosis assistance systems, there is a technical problem that psychiatric medical diagnosis assistance requires a suitable classification paradigm of the subject's mental state in order to effectively classify EEG signals and assist in diagnosis; in the existing psychiatric medical diagnosis assistance systems, there is a technical problem that the EEG data features are complex and diverse, which affects the accuracy of subsequent classification and prediction tasks; in the existing psychiatric medical diagnosis assistance systems, there is a technical problem that complex EEG feature information cannot be simply fused, and multimodal fusion and high-precision classification are required. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a medical diagnosis assistance system based on EEG signals and artificial intelligence classification. In view of the technical problem that in the existing psychiatric medical diagnosis assistance system, psychiatric medical diagnosis assistance requires a suitable classification paradigm of the subject's mental state in order to effectively perform EEG signal classification and diagnosis assistance, this solution creatively constructs an audio-visual mental state paradigm and a resting mental state paradigm, providing good theoretical support for EEG signal acquisition; in view of the technical problem that in the existing psychiatric medical diagnosis assistance system, the EEG data features are complex and diverse, affecting the accuracy of subsequent classification and prediction tasks, this solution creatively adopts The scattered wavelet transform decomposes the EEG signal and calculates the signal feature information, which improves the overall accuracy of the artificial intelligence classification of EEG signals. In view of the technical problem that in the existing psychiatric medical diagnosis auxiliary system, the complex EEG feature information cannot be simply fused and requires multimodal fusion and high-precision classification, this solution creatively adopts the convolutional bidirectional long short-term memory neural network method combined with the multi-head attention mechanism to classify EEG signals. The multi-head attention mechanism concentrates the feature information, uses the convolutional neural subnet to calculate the feature map and uses the bidirectional long short-term memory neural network to classify the signal, which improves the overall automation and usability of the system and provides data and technical support for psychiatric medical diagnosis.

[0005] The technical solution adopted by the present invention is as follows: A medical diagnosis auxiliary system based on EEG signal and artificial intelligence classification provided by the present invention includes an EEG signal acquisition module, a mental state classification paradigm construction module, an EEG information preprocessing module, a feature information processing module, an EEG signal classification module and a medical diagnosis auxiliary module;

[0006] The mental state classification paradigm construction module is used to construct the mental state paradigm of the subject required for the EEG signal acquisition step, obtain the EEG recording acquisition paradigm through the mental state classification paradigm construction, and send the EEG recording acquisition paradigm to the EEG signal acquisition module;

[0007] The EEG signal acquisition module is used to collect EEG signal data required for medical diagnosis assistance, obtain the original EEG signal data by performing signal acquisition and subject data statistics from the EEG acquisition system, and send the original EEG signal data to the EEG information preprocessing module and the EEG signal classification module;

[0008] The EEG information preprocessing module is used to perform preliminary processing on the collected EEG information, and obtain optimized EEG data by using EEGLAB and MATLAB signal processing tools to process the EEG information, and send the optimized EEG data to the feature information processing module and the EEG signal classification module. The EEG information preprocessing includes the following steps: information filtering, data optimization, wavelet decomposition and EEG information preprocessing;

[0009] The feature information processing module is used to perform feature calculation on the EEG data, specifically, to decompose the EEG signals in the optimized EEG data by applying discrete wavelet transform, calculate and construct signal features, obtain EEG feature information data, and send the EEG feature information data to the EEG signal classification module;

[0010] The EEG signal classification module is used to fuse EEG feature information and perform EEG signal classification, specifically using a convolutional bidirectional long short-term memory neural network method combined with a multi-head attention mechanism to classify EEG signals based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data, and send the EEG signal classification data to the medical diagnosis auxiliary module;

[0011] The convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism includes a multi-head attention layer, a convolutional neural network subnet and a bidirectional long short-term memory neural network subnet; the multi-head attention layer is used to concentrate feature information; the convolutional neural network subnet is used to calculate and generate feature maps and perform multi-array feature data processing, specifically including a convolutional layer, a pooling layer and a fully connected layer, wherein the pooling layer specifically adopts multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer to perform multimodal fusion; the bidirectional long short-term memory neural network subnet is used to classify EEG signals, specifically including a forward subnet, a backward subnet and an independent hidden layer;

[0012] The medical diagnosis assistance module is used to assist in medical diagnosis based on the artificial intelligence classification results of the EEG signal, and obtain psychiatric triage reference information through medical diagnosis assistance.

[0013] Furthermore, the mental state classification paradigm is constructed to construct the mental state paradigm of the subject required for the EEG signal acquisition step, specifically, to construct an audio-visual mental state paradigm and a resting mental state paradigm, to obtain an EEG recording acquisition paradigm;

[0014] The audio-visual mental state paradigm includes an auditory stimulation paradigm and a picture stimulation paradigm;

[0015] The auditory stimulation paradigm specifically refers to the EEG acquisition system providing 1000 Hz, 70 dB sinusoidal tone stimulation through a loudspeaker, and the sinusoidal tone stimulation is a paired sound stimulation;

[0016] The paired sounds specifically refer to paired sound sources with an interval of 512 milliseconds and an interval between pairs of 5 seconds, for a total of 36 pairs;

[0017] The paired sound stimulation lasted for 3 minutes, and during the paired sound stimulation, the subjects were required to look at the symbol on the computer screen about 80 cm in front of them;

[0018] The picture stimulation paradigm, the pictures specifically refer to a group of pictures that match the stimulation level, complexity, brightness and color, including emotional pictures and non-emotional pictures;

[0019] The picture stimulation specifically includes 20 pairs of picture stimulations that appear in a random order, each picture is displayed for 5 seconds, the time interval between the display of each two pictures is 10 seconds, and each picture is repeated 3 times;

[0020] The resting state mental state paradigm includes an eyes-open resting state paradigm and an eyes-closed resting state paradigm;

[0021] The resting-state mental state paradigm specifically lasts for 6 minutes, including 3 minutes of eyes-open resting state and 3 minutes of eyes-closed resting state.

[0022] Furthermore, the EEG signal acquisition is used to acquire EEG signal data required for medical diagnosis assistance, specifically, the EEG acquisition system is used to acquire the original EEG signal data through signal acquisition and subject data statistics, and the original EEG signal data includes brain alternating current signal data, subject basic information data and subject cognitive assessment data;

[0023] The EEG acquisition system specifically uses a 64-channel EEG signal amplifier to perform EEG recording on the subject according to the EEG recording acquisition paradigm;

[0024] The EEG recording includes cognitive tasks and resting-state EEG acquisition. The resting-state EEG acquisition specifically refers to continuously recording the brain alternating current signal, and using 24-bit resolution and 1024 Hz sampling rate to digitize and save the EEG signal. The signal range of the brain alternating current signal is 0.03 Hz to 100 Hz. The 64-lead EEG signal amplifier specifically refers to a helmet-type dry electrode EEG cap with 64 dry electrodes.

[0025] Furthermore, the EEG information preprocessing is used to perform preliminary processing on the collected EEG information, specifically using the signal processing tools of EEGLAB and MATLAB to process the EEG information to obtain optimized EEG data, including the following steps:

[0026] Information filtering, specifically using the infomaxICA method to extract eye blink information and filter artifacts;

[0027] Data optimization, specifically data optimization through data denoising;

[0028] Wavelet decomposition, specifically, decomposing EEG signals through wavelet transform;

[0029] The EEG information preprocessing is specifically to perform EEG information preprocessing on the raw EEG signal data through the information filtering, the data optimization and the wavelet decomposition to obtain optimized EEG data.

[0030] Furthermore, the feature information processing is used to perform feature calculation on the EEG data, specifically by applying discrete wavelet transform to decompose the EEG signal in the optimized EEG data, calculate and construct signal features, and obtain EEG feature information data, which specifically includes the following steps:

[0031] Construct contrast features, the calculation formula is:

[0032]

[0033] In the formula, F Contis the contrast feature, which is used to represent the difference between the amplitude values ​​of the instantaneous signal and its adjacent signal in the entire time series, N is the total number of rows and columns of the EEG signal feature vector, i is the row vector column index of the EEG signal feature vector, and X(·) is the EEG signal amplitude function;

[0034] Construct the homogeneity feature, and the calculation formula is:

[0035]

[0036] In the formula, F Hom is the homogeneity feature, which is used to represent the uniformity of the distribution elements of the co-occurrence matrix and its diagonal, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, and X(·) is the EEG signal amplitude function;

[0037] Construct the consistency feature, and the calculation formula is:

[0038]

[0039] In the formula, F Ene is the consistency feature, which is used to represent the sum of the square elements of the co-occurrence matrix, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, and X(·) is the EEG signal amplitude function;

[0040] Construct correlation features, and the calculation formula is:

[0041]

[0042] In the formula, F Cor is a correlation feature, which is used to represent the linear relationship between the row vector and two instantaneous amplitude values. N is the total number of rows of the EEG signal feature vector, i is the row vector column index of the EEG signal feature vector, j is the adjacent instantaneous amplitude vector identifier when the row vector column index is i, μ i is the row vector mean, μ j is the mean of the adjacent instantaneous amplitude vector, σ i is the row vector standard deviation, σ j is the standard deviation of the adjacent instantaneous amplitude vector, X(·) is the EEG signal amplitude function;

[0043] Construct the standard deviation feature, and the calculation formula is:

[0044]

[0045] In the formula, F SDis the standard deviation feature, which is used to represent the mean square change of the instantaneous amplitude value and the average value of the EEG signal, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, X(·) is the EEG signal amplitude function, and μ is the average value of the EEG signal;

[0046] Construct entropy features, and the calculation formula is:

[0047]

[0048] In the formula, F Ent is the entropy feature, which is used to describe the texture features of the row vector, L is the total number of amplitudes in the EEG signal, k is the amplitude index, and P k is the probability corresponding to each k-th amplitude value;

[0049] Construct the median feature, and the calculation formula is:

[0050]

[0051] In the formula, F Mea is the median feature, which is used to represent the central tendency of the probability distribution in the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, and P(·) is the probability distribution of the row vector of the EEG signal feature vector;

[0052] Construct the skewness feature, and the calculation formula is:

[0053]

[0054] In the formula, F Ske is the skewness feature, which is used to indicate the degree of asymmetry of the probability distribution in the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, P(·) is the probability distribution function of the row vector of the EEG signal feature vector, μ is the mean value of the EEG signal, and σ is the standard deviation of the EEG signal;

[0055] Construct the smoothness feature, and the calculation formula is:

[0056]

[0057] In the formula, F Smt is the smoothness feature, used to measure the contrast of amplitude values, and σ is the standard deviation of the EEG signal;

[0058] Construct the kurtosis feature, and the calculation formula is:

[0059]

[0060] In the formula, F Kuris the peak feature, used to represent the normal distribution of the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the index of the row vector column of the EEG signal feature vector, P(·) is the probability distribution function of the row vector of the EEG signal feature vector, μ is the mean value of the EEG signal, and σ is the standard deviation of the EEG signal;

[0061] Construct variance features, and the calculation formula is:

[0062] F Var =F SD 2 ;

[0063] In the formula, F Var is the variance feature, F SD is the standard deviation characteristic;

[0064] Construct the reverse motion feature, and the calculation formula is:

[0065]

[0066] Where FIDM is the inverse difference motion feature, which is used to represent the texture level of the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, and P(·) is the probability distribution function of the row vector of the EEG signal feature vector;

[0067] Constructing signal features, specifically by constructing contrast features, constructing homogeneity features, constructing consistency features, constructing correlation features, constructing standard deviation features, constructing entropy features, constructing median features, constructing skewness features, constructing smoothness features, constructing kurtosis features, constructing variance features and constructing inverse motion features, constructing signal features to obtain EEG feature information data.

[0068] Furthermore, the EEG signal classification is used to fuse EEG feature information and perform EEG signal classification, specifically using a convolutional bidirectional long short-term memory neural network method combined with a multi-head attention mechanism to perform EEG signal classification based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data;

[0069] The convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism includes a multi-head attention layer, a convolutional neural network subnet and a bidirectional long short-term memory neural network subnet;

[0070] The multi-head attention layer is used to concentrate feature information;

[0071] The convolutional neural network subnet is used to calculate and generate feature maps and perform multi-array feature data processing, and specifically includes a convolution layer, a pooling layer, and a fully connected layer. The pooling layer specifically uses multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer to perform multimodal fusion;

[0072] The bidirectional long short-term memory neural network subnet is used for EEG signal classification, and specifically includes a forward subnet, a backward subnet and an independent hidden layer;

[0073] The method of using a convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism to classify EEG signals based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data includes:

[0074] Training data construction, specifically, randomly extracting 70% of the data from the optimized EEG data as training data samples;

[0075] Constructing a multi-head attention layer includes the following steps:

[0076] Construct the attention matrix, the calculation formula is:

[0077]

[0078] Where Att(Q,K,V) is the calculated value of the attention matrix, softmax(·) is the softmax classifier function, Q is the query vector, K is the key vector, V is the value vector, T is the transpose operator, and d k is the characteristic dimension of the EEG characteristic information data;

[0079] Construct a parallel attention head, and the calculation formula is:

[0080]

[0081] In the formula, head i′ is a parallel attention head used to capture information from multiple subspaces, i′ is the parallel attention head index, Att(·) is the calculation function of the attention matrix, Q is the query vector, K is the key vector, V is the value vector, is the query prediction weight matrix, is the key prediction weight matrix, is the value prediction weight matrix;

[0082] Construct a multi-head attention mechanism, the calculation formula is:

[0083] MultiHead(Q,K,V)=Concat(head , ...,head , ) O ;

[0084] Where MultiHead(Q, K, V) is the calculated value of the multi-head attention mechanism, Q is the query vector, K is the key vector, V is the value vector, Concat(·) is the concatenation operation function, head1 is the first parallel attention head, and head h is the hth parallel attention head, where h is the total number of parallel attention heads, and W O is the linear output function weight matrix;

[0085] Constructing a convolutional neural network subnet includes the following steps:

[0086] Constructing a convolutional layer, specifically, by calculating a feature map and activating it with a nonlinear activation function, to construct the convolutional layer, the calculation formula of the convolutional layer is:

[0087] I = ReLU(w*X′+b);

[0088] Where I is the output feature map of the convolutional layer, ReLU(·) is the nonlinear activation function, w is the weight of the convolutional layer, X′ is the input sample of the convolutional layer, and b is the bias term of the convolutional layer;

[0089] Constructing a pooling layer, specifically using multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer, performing multimodal fusion, and constructing the pooling layer;

[0090] The multimodal low-rank bilinear pooling step includes: feature extraction, principal component analysis feature dimensionality reduction and bilinear pooling to obtain a multimodal fusion feature map;

[0091] Constructing a fully connected layer, specifically receiving the multimodal fusion feature map obtained by the multimodal low-rank bilinear pooling, and connecting the multimodal fusion feature map to a bidirectional long short-term memory neural network subnet;

[0092] Constructing a bidirectional long short-term memory neural network subnetwork includes the following steps:

[0093] Construct a forward subnet, specifically, by constructing a forward input gate, a forward forget gate, and a forward output gate to obtain a forward long short-term memory neural network;

[0094] Constructing a backward subnet, specifically, by constructing a backward input gate, a backward forget gate, and a backward output gate to obtain a backward long short-term memory neural network;

[0095] Constructing an independent hidden layer for capturing bidirectional information of the sample, specifically by constructing an independent hidden layer before the input layer of the forward subnet and the backward subnet, and inputting the hidden layer outputs into the forward input gate and the backward input gate, respectively, to construct the independent hidden layer;

[0096] EEG signal classification model training, specifically, training the EEG signal classification model through the training data construction, the construction of the multi-head attention layer, the construction of the convolutional neural network subnet and the construction of the bidirectional long short-term memory neural network subnet, to obtain the EEG signal classification artificial intelligence model Model AI ;

[0097] EEG signal classification, specifically by using the EEG signal classification artificial intelligence model Model AI , performing EEG signal classification on the EEG signal raw data to obtain EEG signal classification data.

[0098] Furthermore, the medical diagnosis assistance is used to perform medical diagnosis assistance based on the artificial intelligence classification results of the EEG signal, specifically to perform psychiatric medical diagnosis assistance based on the EEG signal classification data to obtain a diagnosis assistance report. The diagnosis assistance report is generated based on the EEG signal classification data, specifically referring to various psychiatric data features in the EEG of a single patient, and the psychiatric data features include depressive components, anxiety components, schizophrenia components, affective disorder components and Alzheimer's disease components.

[0099] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0100] (1) In view of the technical problem that the existing psychiatric medical diagnosis assistance system requires a suitable classification paradigm of the subject's mental state in order to effectively classify EEG signals and assist in diagnosis, this solution creatively constructs an audio-visual mental state paradigm and a resting mental state paradigm, providing good theoretical support for EEG signal acquisition;

[0101] (2) In view of the technical problem that the EEG data features in the existing psychiatric medical diagnosis assistance system are complex and diverse, which affects the accuracy of subsequent classification and prediction tasks, this solution creatively uses discrete wavelet transform to decompose EEG signals and calculate signal feature information, thereby improving the overall accuracy of EEG signal artificial intelligence classification;

[0102] (3) In order to address the technical problem that in existing psychiatric medical diagnosis assistance systems, complex EEG feature information cannot be simply fused and requires multimodal fusion and high-precision classification, this solution creatively uses a convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism to classify EEG signals. The multi-head attention mechanism is used to concentrate feature information, a convolutional neural subnet is used to calculate feature maps, and a bidirectional long short-term memory neural network is used to classify signals, thereby improving the overall automation and usability of the system and providing data and technical support for psychiatric medical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 A schematic diagram of the structure of a medical diagnosis assistance system based on EEG signals and artificial intelligence classification provided by the present invention;

[0104] Figure 2 A flowchart of the steps performed by a medical diagnosis aid based on EEG signals and artificial intelligence classification provided by the present invention;

[0105] Figure 3 A data flow diagram of a medical diagnosis assistance system based on EEG signals and artificial intelligence classification provided by the present invention;

[0106] Figure 4 A flowchart of the steps performed by the EEG information preprocessing module;

[0107] Figure 5 A flowchart of the steps performed by the feature information processing module;

[0108] Figure 6 Schematic diagram of the flow of steps performed by the EEG signal classification module.

[0109] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0110] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0111] Example 1, see Figure 1 and Figure 3 , a medical diagnosis auxiliary system based on EEG signal and artificial intelligence classification provided by the present invention includes an EEG signal acquisition module, a mental state classification paradigm construction module, an EEG information preprocessing module, a feature information processing module, an EEG signal classification module and a medical diagnosis auxiliary module;

[0112] The mental state classification paradigm construction module is used to construct the mental state paradigm of the subject required for the EEG signal acquisition step, obtain the EEG recording acquisition paradigm through the mental state classification paradigm construction, and send the EEG recording acquisition paradigm to the EEG signal acquisition module;

[0113] The EEG signal acquisition module is used to collect EEG signal data required for medical diagnosis assistance, obtain the original EEG signal data by performing signal acquisition and subject data statistics from the EEG acquisition system, and send the original EEG signal data to the EEG information preprocessing module and the EEG signal classification module;

[0114] The EEG information preprocessing module is used to perform preliminary processing on the collected EEG information, and obtain optimized EEG data by using EEGLAB and MATLAB signal processing tools to process the EEG information, and send the optimized EEG data to the feature information processing module and the EEG signal classification module. The EEG information preprocessing includes the following steps: information filtering, data optimization, wavelet decomposition and EEG information preprocessing;

[0115] The feature information processing module is used to perform feature calculation on the EEG data, specifically, to decompose the EEG signals in the optimized EEG data by applying discrete wavelet transform, calculate and construct signal features, obtain EEG feature information data, and send the EEG feature information data to the EEG signal classification module;

[0116] The EEG signal classification module is used to fuse EEG feature information and perform EEG signal classification, specifically using a convolutional bidirectional long short-term memory neural network method combined with a multi-head attention mechanism to classify EEG signals based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data, and send the EEG signal classification data to the medical diagnosis auxiliary module;

[0117] The convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism includes a multi-head attention layer, a convolutional neural network subnet and a bidirectional long short-term memory neural network subnet; the multi-head attention layer is used to concentrate feature information; the convolutional neural network subnet is used to calculate and generate feature maps and perform multi-array feature data processing, specifically including a convolutional layer, a pooling layer and a fully connected layer, wherein the pooling layer specifically adopts multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer to perform multimodal fusion; the bidirectional long short-term memory neural network subnet is used to classify EEG signals, specifically including a forward subnet, a backward subnet and an independent hidden layer;

[0118] The medical diagnosis assistance module is used to assist in medical diagnosis based on the artificial intelligence classification results of the EEG signal, and obtain psychiatric triage reference information through medical diagnosis assistance.

[0119] Embodiment 2, this embodiment is based on the above embodiment, see Figure 1 , Figure 2 and Figure 3The mental state classification paradigm is constructed to construct the mental state paradigm of the subject required for the EEG signal acquisition step, specifically, to construct an audio-visual mental state paradigm and a resting mental state paradigm, to obtain an EEG recording acquisition paradigm;

[0120] The audio-visual mental state paradigm includes an auditory stimulation paradigm and a picture stimulation paradigm;

[0121] The auditory stimulation paradigm specifically refers to the EEG acquisition system providing 1000 Hz, 70 dB sinusoidal tone stimulation through a loudspeaker, and the sinusoidal tone stimulation is a paired sound stimulation;

[0122] The paired sounds specifically refer to paired sound sources with an interval of 512 milliseconds and an interval between pairs of 5 seconds, for a total of 36 pairs;

[0123] The paired sound stimulation lasted for 3 minutes, and during the paired sound stimulation, the subjects were required to look at the symbol on the computer screen about 80 cm in front of them;

[0124] The picture stimulation paradigm, the pictures specifically refer to a group of pictures that match the stimulation level, complexity, brightness and color, including emotional pictures and non-emotional pictures;

[0125] The picture stimulation specifically includes 20 pairs of picture stimulations that appear in a random order, each picture is displayed for 5 seconds, the time interval between the display of each two pictures is 10 seconds, and each picture is repeated 3 times;

[0126] The resting state mental state paradigm includes an eyes-open resting state paradigm and an eyes-closed resting state paradigm;

[0127] The resting-state mental state paradigm specifically lasts for 6 minutes, including 3 minutes of eyes-open resting state and 3 minutes of eyes-closed resting state.

[0128] By executing the above operations, in order to effectively classify EEG signals and provide diagnostic assistance, the existing psychiatric medical diagnosis assistance system requires a suitable classification paradigm of the subject's mental state. This solution creatively constructs an audio-visual mental state paradigm and a resting mental state paradigm, providing good theoretical support for EEG signal acquisition.

[0129] Embodiment 3, this embodiment is based on the above embodiment, see Figure 1 , Figure 2 and Figure 3The EEG signal acquisition is used to acquire EEG signal data required for medical diagnosis assistance, specifically, the EEG acquisition system is used to acquire the original EEG signal data through signal acquisition and subject data statistics, and the original EEG signal data includes brain alternating current signal data, subject basic information data and subject cognitive assessment data;

[0130] The EEG acquisition system specifically uses a 64-channel EEG signal amplifier to perform EEG recording on the subject according to the EEG recording acquisition paradigm;

[0131] The EEG recording includes cognitive tasks and resting-state EEG acquisition. The resting-state EEG acquisition specifically refers to continuously recording the brain alternating current signal, and using 24-bit resolution and 1024 Hz sampling rate to digitize and save the EEG signal. The signal range of the brain alternating current signal is 0.03 Hz to 100 Hz. The 64-lead EEG signal amplifier specifically refers to a helmet-type dry electrode EEG cap with 64 dry electrodes.

[0132] Embodiment 4, this embodiment is based on the above embodiment, see Figure 2 , Figure 3 and Figure 4 The EEG information preprocessing is used to preliminarily process the collected EEG information, specifically using the signal processing tools of EEGLAB and MATLAB to process the EEG information to obtain optimized EEG data, including the following steps:

[0133] Information filtering, specifically using the infomaxICA method to extract eye blink information and filter artifacts;

[0134] Data optimization, specifically data optimization through data denoising;

[0135] Wavelet decomposition, specifically, decomposing EEG signals through wavelet transform;

[0136] The EEG information preprocessing is specifically to perform EEG information preprocessing on the raw EEG signal data through the information filtering, the data optimization and the wavelet decomposition to obtain optimized EEG data.

[0137] Embodiment 5, this embodiment is based on the above embodiment, see Figure 1 , Figure 2 and Figure 5 The feature information processing is used to perform feature calculation on the EEG data, specifically by applying discrete wavelet transform to decompose the EEG signal in the optimized EEG data, calculate and construct signal features, and obtain EEG feature information data, specifically including the following steps:

[0138] Construct contrast features, and the calculation formula is:

[0139]

[0140] In the formula, F Cont is the contrast feature, which is used to represent the difference between the amplitude values ​​of the instantaneous signal and its adjacent signal in the entire time series, N is the total number of rows and columns of the EEG signal feature vector, i is the row vector column index of the EEG signal feature vector, and X(·) is the EEG signal amplitude function;

[0141] Construct the homogeneity feature, and the calculation formula is:

[0142]

[0143] In the formula, F Hom is the homogeneity feature, which is used to represent the uniformity of the distribution elements of the co-occurrence matrix and its diagonal, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, and X(·) is the EEG signal amplitude function;

[0144] Construct the consistency feature, and the calculation formula is:

[0145]

[0146] In the formula, F Ene is the consistency feature, which is used to represent the sum of the square elements of the co-occurrence matrix, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, and X(·) is the EEG signal amplitude function;

[0147] Construct correlation features, and the calculation formula is:

[0148]

[0149] In the formula, F Cor is a correlation feature, which is used to represent the linear relationship between the row vector and two instantaneous amplitude values. N is the total number of rows of the EEG signal feature vector, i is the row vector column index of the EEG signal feature vector, j is the adjacent instantaneous amplitude vector identifier when the row vector column index is i, μ i is the row vector mean, μ j is the mean of the adjacent instantaneous amplitude vector, σ i is the row vector standard deviation, σ j is the standard deviation of the adjacent instantaneous amplitude vector, X(·) is the EEG signal amplitude function;

[0150] Construct the standard deviation feature, and the calculation formula is:

[0151]

[0152] In the formula, F SDis the standard deviation feature, which is used to represent the mean square change of the instantaneous amplitude value and the average value of the EEG signal, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, X(·) is the EEG signal amplitude function, and μ is the average value of the EEG signal;

[0153] Construct entropy features, and the calculation formula is:

[0154]

[0155] In the formula, F Ent is the entropy feature, which is used to describe the texture features of the row vector, L is the total number of amplitudes in the EEG signal, k is the amplitude index, and P k is the probability corresponding to each k-th amplitude value;

[0156] Construct the median feature, and the calculation formula is:

[0157]

[0158] In the formula, F Mea is the median feature, which is used to represent the central tendency of the probability distribution in the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, and P(·) is the probability distribution of the row vector of the EEG signal feature vector;

[0159] Construct the skewness feature, and the calculation formula is:

[0160]

[0161] In the formula, F Ske is the skewness feature, which is used to indicate the degree of asymmetry of the probability distribution in the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, P(·) is the probability distribution function of the row vector of the EEG signal feature vector, μ is the mean value of the EEG signal, and σ is the standard deviation of the EEG signal;

[0162] Construct the smoothness feature, and the calculation formula is:

[0163]

[0164] In the formula, F Smt is the smoothness feature, used to measure the contrast of amplitude values, and σ is the standard deviation of the EEG signal;

[0165] Construct the kurtosis feature, and the calculation formula is:

[0166]

[0167] In the formula, F Kuris the peak feature, used to represent the normal distribution of the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the index of the row vector column of the EEG signal feature vector, P(·) is the probability distribution function of the row vector of the EEG signal feature vector, μ is the mean value of the EEG signal, and σ is the standard deviation of the EEG signal;

[0168] Construct variance features, and the calculation formula is:

[0169] F Var =F SD 2 ;

[0170] In the formula, F Var is the variance feature, F SD is the standard deviation characteristic;

[0171] Construct the reverse motion feature, and the calculation formula is:

[0172]

[0173] Where FIDM is the inverse difference motion feature, which is used to represent the texture level of the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, and P(·) is the probability distribution function of the row vector of the EEG signal feature vector;

[0174] Constructing signal features, specifically by constructing contrast features, constructing homogeneity features, constructing consistency features, constructing correlation features, constructing standard deviation features, constructing entropy features, constructing median features, constructing skewness features, constructing smoothness features, constructing kurtosis features, constructing variance features and constructing inverse motion features, constructing signal features to obtain EEG feature information data.

[0175] By performing the above operations, this solution creatively uses discrete wavelet transforms to decompose EEG signals and calculate signal feature information in order to address the technical problem that in existing psychiatric medical diagnosis assistance systems, EEG data features are complex and diverse, which affects the accuracy of subsequent classification and prediction tasks. This improves the overall accuracy of EEG signal artificial intelligence classification.

[0176] Embodiment 6, this embodiment is based on the above embodiment, see Figure 1 , Figure 2 and Figure 6 The EEG signal classification is used to fuse EEG feature information and perform EEG signal classification, specifically using a convolutional bidirectional long short-term memory neural network method combined with a multi-head attention mechanism to perform EEG signal classification based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data;

[0177] The convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism includes a multi-head attention layer, a convolutional neural network subnet and a bidirectional long short-term memory neural network subnet;

[0178] The multi-head attention layer is used to concentrate feature information;

[0179] The convolutional neural network subnet is used to calculate and generate feature maps and perform multi-array feature data processing, and specifically includes a convolution layer, a pooling layer, and a fully connected layer. The pooling layer specifically uses multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer to perform multimodal fusion;

[0180] The bidirectional long short-term memory neural network subnet is used for EEG signal classification, and specifically includes a forward subnet, a backward subnet and an independent hidden layer;

[0181] The method of using a convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism to classify EEG signals based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data includes:

[0182] Training data construction, specifically, randomly extracting 70% of the data from the optimized EEG data as training data samples;

[0183] Constructing a multi-head attention layer includes the following steps:

[0184] Construct the attention matrix, the calculation formula is:

[0185]

[0186] Where Att(Q,K,V) is the calculated value of the attention matrix, softmax(·) is the softmax classifier function, Q is the query vector, K is the key vector, V is the value vector, T is the transpose operator, and d k is the characteristic dimension of the EEG characteristic information data;

[0187] Construct a parallel attention head, and the calculation formula is:

[0188]

[0189] In the formula, head i′ is a parallel attention head used to capture information from multiple subspaces, i′ is the parallel attention head index, Att(·) is the calculation function of the attention matrix, Q is the query vector, K is the key vector, V is the value vector, is the query prediction weight matrix, is the key prediction weight matrix, is the value prediction weight matrix;

[0190] Construct a multi-head attention mechanism, the calculation formula is:

[0191] MultiHead(Q,K,V)=Concat(head1,..,head h )W O ;

[0192] Where MultiHead(Q, K, V) is the calculated value of the multi-head attention mechanism, Q is the query vector, K is the key vector, V is the value vector, Concat(·) is the concatenation operation function, head1 is the first parallel attention head, and head h is the hth parallel attention head, where h is the total number of parallel attention heads, and W O is the linear output function weight matrix;

[0193] Constructing a convolutional neural network subnet includes the following steps:

[0194] Constructing a convolutional layer, specifically, by calculating a feature map and activating it with a nonlinear activation function, to construct the convolutional layer, the calculation formula of the convolutional layer is:

[0195] I = ReLU(w*X′+b);

[0196] Where I is the output feature map of the convolutional layer, ReLU(·) is the nonlinear activation function, w is the weight of the convolutional layer, X is the input sample of the convolutional layer, and b is the bias term of the convolutional layer;

[0197] Constructing a pooling layer, specifically using multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer, performing multimodal fusion, and constructing the pooling layer;

[0198] The multimodal low-rank bilinear pooling step includes: feature extraction, principal component analysis feature dimensionality reduction and bilinear pooling to obtain a multimodal fusion feature map;

[0199] Constructing a fully connected layer, specifically receiving the multimodal fusion feature map obtained by the multimodal low-rank bilinear pooling, and connecting the multimodal fusion feature map to a bidirectional long short-term memory neural network subnet;

[0200] Constructing a bidirectional long short-term memory neural network subnetwork includes the following steps:

[0201] Construct a forward subnet, specifically, by constructing a forward input gate, a forward forget gate, and a forward output gate to obtain a forward long short-term memory neural network;

[0202] Constructing a backward subnet, specifically, by constructing a backward input gate, a backward forget gate, and a backward output gate to obtain a backward long short-term memory neural network;

[0203] Constructing an independent hidden layer for capturing bidirectional information of the sample, specifically by constructing an independent hidden layer before the input layer of the forward subnet and the backward subnet, and inputting the hidden layer outputs into the forward input gate and the backward input gate, respectively, to construct the independent hidden layer;

[0204] EEG signal classification model training, specifically, training the EEG signal classification model through the training data construction, the construction of the multi-head attention layer, the construction of the convolutional neural network subnet and the construction of the bidirectional long short-term memory neural network subnet, to obtain the EEG signal classification artificial intelligence model Model AI ;

[0205] EEG signal classification, specifically by using the EEG signal classification artificial intelligence model Model AI , performing EEG signal classification on the EEG signal raw data to obtain EEG signal classification data.

[0206] By executing the above operations, in order to solve the technical problem that in the existing psychiatric medical diagnosis auxiliary system, complex EEG feature information cannot be simply fused and multimodal fusion and high-precision classification are required, this solution creatively adopts the method of convolutional bidirectional long short-term memory neural network combined with multi-head attention mechanism to classify EEG signals. It concentrates feature information through the multi-head attention mechanism, uses convolutional neural subnet to calculate feature maps and uses bidirectional long short-term memory neural network to classify signals, which improves the overall automation and usability of the system and provides data and technical support for psychiatric medical diagnosis.

[0207] Embodiment 7, this embodiment is based on the above embodiment, see Figure 1 , Figure 2 and Figure 3 The medical diagnosis assistance is used to perform medical diagnosis assistance based on the artificial intelligence classification results of the EEG signal, specifically to perform psychiatric medical diagnosis assistance based on the EEG signal classification data to obtain a diagnosis assistance report. The diagnosis assistance report is generated based on the EEG signal classification data, specifically referring to the psychiatric data features in the EEG of a single patient, and the psychiatric data features include depression components, anxiety components, schizophrenia components, affective disorder components and Alzheimer's disease components.

[0208] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process and method including a series of elements include not only those elements, but also other elements not explicitly listed, or also include elements inherent to such process and method.

[0209] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

[0210] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A medical diagnosis assistance system based on EEG signals and artificial intelligence classification, characterized by: It includes EEG signal acquisition module, mental state classification paradigm construction module, EEG information preprocessing module, feature information processing module, EEG signal classification module and medical diagnosis auxiliary module; The mental state classification paradigm construction module is used to construct the mental state paradigm of the subject required for the EEG signal acquisition step, obtain the EEG recording acquisition paradigm through the mental state classification paradigm construction, and send the EEG recording acquisition paradigm to the EEG signal acquisition module; The mental state paradigm of the subject required for the EEG signal acquisition step specifically refers to constructing an audio-visual mental state paradigm and constructing a resting mental state paradigm; The audio-visual mental state paradigm includes an auditory stimulation paradigm and a picture stimulation paradigm; The auditory stimulation paradigm specifically refers to the EEG acquisition system providing 1000 Hz, 70 dB sinusoidal tone stimulation through a loudspeaker, and the sinusoidal tone stimulation is a paired sound stimulation; The paired sounds specifically refer to paired sound sources with an interval of 512 milliseconds and an interval between pairs of 5 seconds, for a total of 36 pairs; The paired sound stimulation lasted for 3 minutes, and during the paired sound stimulation, the subjects were required to look at the symbol on the computer screen about 80 cm in front of them; The picture stimulation paradigm, the pictures specifically refer to a group of pictures that match the stimulation level, complexity, brightness and color, including emotional pictures and non-emotional pictures; The picture stimulation specifically includes 20 pairs of picture stimulations that appear in a random order, each picture is displayed for 5 seconds, the interval between the display of each two pictures is 10 seconds, and each picture is repeated 3 times; The resting state mental state paradigm includes an eyes-open resting state paradigm and an eyes-closed resting state paradigm; The resting state mental state paradigm specifically lasts for 6 minutes, including 3 minutes of eyes-open resting state and 3 minutes of eyes-closed resting state; The EEG signal acquisition module is used to collect EEG signal data required for medical diagnosis assistance, obtain the original EEG signal data by performing signal acquisition and subject data statistics from the EEG acquisition system, and send the original EEG signal data to the EEG information preprocessing module and the EEG signal classification module; The EEG information preprocessing module is used to perform preliminary processing on the collected EEG information, and obtain optimized EEG data by using EEGLAB and MATLAB signal processing tools to process the EEG information, and send the optimized EEG data to the feature information processing module and the EEG signal classification module. The EEG information preprocessing includes the following steps: information filtering, data optimization, wavelet decomposition and EEG information preprocessing; The feature information processing module is used to perform feature calculation on the EEG data, specifically, to decompose the EEG signals in the optimized EEG data by applying discrete wavelet transform, calculate and construct signal features, obtain EEG feature information data, and send the EEG feature information data to the EEG signal classification module; The method of decomposing the EEG signal in the optimized EEG data by applying discrete wavelet transform, calculating and constructing signal features, and obtaining EEG feature information data specifically includes the following steps: constructing contrast features, constructing homogeneity features, constructing consistency features, constructing correlation features, constructing standard deviation features, constructing entropy features, constructing median features, constructing skewness features, constructing smoothness features, constructing kurtosis features, constructing variance features, constructing inverse motion features, and constructing signal features; The EEG signal classification module is used to fuse EEG feature information and perform EEG signal classification, specifically using a convolutional bidirectional long short-term memory neural network method combined with a multi-head attention mechanism to classify EEG signals based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data, and send the EEG signal classification data to the medical diagnosis auxiliary module; The convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism includes a multi-head attention layer, a convolutional neural network subnet and a bidirectional long short-term memory neural network subnet; the multi-head attention layer is used to concentrate feature information; the convolutional neural network subnet is used to calculate and generate feature maps and perform multi-array feature data processing, specifically including a convolutional layer, a pooling layer and a fully connected layer, wherein the pooling layer specifically adopts multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer to perform multimodal fusion; the bidirectional long short-term memory neural network subnet is used to classify EEG signals, specifically including a forward subnet, a backward subnet and an independent hidden layer; The medical diagnosis assistance module is used to assist in medical diagnosis based on the artificial intelligence classification results of the EEG signal, and obtain psychiatric triage reference information through medical diagnosis assistance.

2. A medical diagnosis assistance system based on EEG signals and artificial intelligence classification according to claim 1, characterized in that: The contrast feature is constructed and the calculation formula is: In the formula, F Cont is the contrast feature, which is used to represent the difference between the amplitude values ​​of the instantaneous signal and its adjacent signal in the entire time series, N is the total number of rows and columns of the EEG signal feature vector, i is the row vector column index of the EEG signal feature vector, and X(·) is the EEG signal amplitude function; The construction homogeneity feature is calculated by the formula: In the formula, F Hom is the homogeneity feature, which is used to represent the uniformity of the distribution elements of the co-occurrence matrix and its diagonal, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, and X(·) is the EEG signal amplitude function; The calculation formula for constructing the consistency feature is: In the formula, F Ene is the consistency feature, which is used to represent the sum of the square elements of the co-occurrence matrix, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, and X(·) is the EEG signal amplitude function; The calculation formula for constructing the correlation feature is: In the formula, F Cor is a correlation feature, which is used to represent the linear relationship between the row vector and two instantaneous amplitude values. N is the total number of rows of the EEG signal feature vector, i is the row vector column index of the EEG signal feature vector, j is the adjacent instantaneous amplitude vector identifier when the row vector column index is i, μ i is the row vector mean, μ j is the mean of the adjacent instantaneous amplitude vector, σ i is the row vector standard deviation, σ j is the standard deviation of the adjacent instantaneous amplitude vector, X(·) is the EEG signal amplitude function; The calculation formula for constructing the standard deviation feature is: In the formula, F SD is the standard deviation feature, which is used to represent the mean square change of the instantaneous amplitude value and the average value of the EEG signal, N is the total number of rows and columns of the EEG signal feature vector, i is the index of the row vector column number of the EEG signal feature vector, X(·) is the EEG signal amplitude function, and μ is the average value of the EEG signal; The entropy value feature is constructed and the calculation formula is: In the formula, F Ent is the entropy feature, which is used to describe the texture features of the row vector, L is the total number of amplitudes in the EEG signal, k is the amplitude index, and P k is the probability corresponding to each k-th amplitude value; The median feature is constructed and the calculation formula is: In the formula, F Mea is the median feature, which is used to represent the central tendency of the probability distribution in the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, and P(·) is the probability distribution of the row vector of the EEG signal feature vector; The constructed skewness feature is calculated as follows: In the formula, F Ske is the skewness feature, which is used to indicate the degree of asymmetry of the probability distribution in the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the column index of the row vector of the EEG signal feature vector, P(·) is the probability distribution function of the row vector of the EEG signal feature vector, μ is the mean value of the EEG signal, and σ is the standard deviation of the EEG signal; The calculation formula for constructing the smoothness feature is: In the formula, F Smt is the smoothness feature, used to measure the contrast of amplitude values, and σ is the standard deviation of the EEG signal; The constructed kurtosis feature is calculated as follows: In the formula, F Kur is the peak feature, used to represent the normal distribution of the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the index of the row vector column of the EEG signal feature vector, P(·) is the probability distribution function of the row vector of the EEG signal feature vector, μ is the mean value of the EEG signal, and σ is the standard deviation of the EEG signal; The variance feature is constructed and the calculation formula is: F Var =F SD 2 ; In the formula, F Var is the variance feature, F SD is the standard deviation characteristic; The calculation formula for constructing the reverse difference motion feature is: In the formula, F IDM is the inverse difference motion feature, used to represent the texture level of the row vector, N is the total number of columns of the row vector of the EEG signal feature vector, i is the index of the row vector column of the EEG signal feature vector, and P(·) is the probability distribution function of the row vector of the EEG signal feature vector; The constructed signal features are specifically constructed by constructing contrast features, constructing homogeneity features, constructing consistency features, constructing correlation features, constructing standard deviation features, constructing entropy features, constructing median features, constructing skewness features, constructing smoothness features, constructing kurtosis features, constructing variance features and constructing inverse motion features to construct signal features and obtain EEG feature information data.

3. The medical diagnosis auxiliary system based on EEG signal and artificial intelligence classification according to claim 2, characterized in that: In the EEG signal classification, the convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism includes a multi-head attention layer, a convolutional neural network subnet and a bidirectional long short-term memory neural network subnet; The multi-head attention layer is used to concentrate feature information; The convolutional neural network subnet is used to calculate and generate feature maps and perform multi-array feature data processing, and specifically includes a convolution layer, a pooling layer, and a fully connected layer. The pooling layer specifically uses multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer to perform multimodal fusion; The bidirectional long short-term memory neural network subnet is used for classifying EEG signals, and specifically includes a forward subnet, a backward subnet and an independent hidden layer.

4. The medical diagnosis auxiliary system based on EEG signal and artificial intelligence classification according to claim 3, characterized in that: In the EEG signal classification, the method of using a convolutional bidirectional long short-term memory neural network combined with a multi-head attention mechanism to classify EEG signals based on the optimized EEG data and the EEG feature information data to obtain EEG signal classification data includes: Training data construction, specifically, randomly extracting 70% of the data from the optimized EEG data as training data samples; Constructing a multi-head attention layer includes the following steps: Construct the attention matrix, the calculation formula is: Where Att(Q,K,V) is the calculated value of the attention matrix, softmax(·) is the softmax classifier function, Q is the query vector, K is the key vector, V is the value vector, T is the transpose operator, and d k is the characteristic dimension of the EEG characteristic information data; Construct a parallel attention head, and the calculation formula is: In the formula, head i′ is a parallel attention head used to capture information from multiple subspaces, i′ is the parallel attention head index, Att(·) is the calculation function of the attention matrix, Q is the query vector, K is the key vector, V is the value vector, is the query prediction weight matrix, is the key prediction weight matrix, is the value prediction weight matrix; Construct a multi-head attention mechanism, the calculation formula is: MultiHead(Q,K,V)=Concat(head1,...,head h )W o ; Where MultiHead(Q, K, V) is the calculated value of the multi-head attention mechanism, Q is the query vector, K is the key vector, V is the value vector, Concat(·) is the concatenation operation function, head1 is the first parallel attention head, and head h is the hth parallel attention head, where h is the total number of parallel attention heads, and W O is the linear output function weight matrix; Constructing a convolutional neural network subnet includes the following steps: Constructing a convolutional layer, specifically, by calculating a feature map and activating it with a nonlinear activation function, to construct the convolutional layer, the calculation formula of the convolutional layer is: I = ReLU(w*X′+b); Where I is the output feature map of the convolutional layer, ReLU(·) is the nonlinear activation function, w is the weight of the convolutional layer, X′ is the input sample of the convolutional layer, and b is the bias term of the convolutional layer; Constructing a pooling layer, specifically using multimodal low-rank bilinear pooling to replace the feature splicing method of the traditional pooling layer, performing multimodal fusion, and constructing the pooling layer; The multimodal low-rank bilinear pooling step includes: feature extraction, principal component analysis feature dimensionality reduction and bilinear pooling to obtain a multimodal fusion feature map; Constructing a fully connected layer, specifically receiving the multimodal fusion feature map obtained by the multimodal low-rank bilinear pooling, and connecting the multimodal fusion feature map to a bidirectional long short-term memory neural network subnet; Constructing a bidirectional long short-term memory neural network subnetwork includes the following steps: Construct a forward subnet, specifically, by constructing a forward input gate, a forward forget gate, and a forward output gate to obtain a forward long short-term memory neural network; Constructing a backward subnet, specifically, by constructing a backward input gate, a backward forget gate, and a backward output gate to obtain a backward long short-term memory neural network; Constructing an independent hidden layer for capturing bidirectional information of the sample, specifically by constructing an independent hidden layer before the input layer of the forward subnet and the backward subnet, and inputting the hidden layer outputs into the forward input gate and the backward input gate, respectively, to construct the independent hidden layer; EEG signal classification model training, specifically, training the EEG signal classification model through the training data construction, the construction of the multi-head attention layer, the construction of the convolutional neural network subnet and the construction of the bidirectional long short-term memory neural network subnet, to obtain the EEG signal classification artificial intelligence model Model AI ; EEG signal classification, specifically by using the EEG signal classification artificial intelligence model Model AI , performing EEG signal classification on the EEG signal raw data to obtain EEG signal classification data.

5. The medical diagnosis auxiliary system based on EEG signal and artificial intelligence classification according to claim 4, characterized in that: In the EEG information preprocessing, the information filtering is specifically to extract blink information and filter artifacts using the infomax_ICA method; The data optimization is specifically to optimize the data by data denoising; The wavelet decomposition is specifically to decompose the EEG signal by wavelet transform; The EEG information preprocessing is specifically to perform EEG information preprocessing on the original EEG signal data through the information filtering, the data optimization and the wavelet decomposition to obtain optimized EEG data.

6. The medical diagnosis auxiliary system based on EEG signal and artificial intelligence classification according to claim 5, characterized in that: The EEG signal acquisition is specifically to obtain the original EEG signal data by using an EEG acquisition system through signal acquisition and subject data statistics, wherein the original EEG signal data includes brain alternating current signal data, subject basic information data and subject cognitive assessment data; The EEG acquisition system specifically uses a 64-channel EEG signal amplifier to perform EEG recording on the subject according to the EEG recording acquisition paradigm; The EEG recording includes cognitive tasks and resting-state EEG acquisition. The resting-state EEG acquisition specifically refers to continuously recording the brain alternating current signal, and using 24-bit resolution and 1024 Hz sampling rate to digitize and save the EEG signal. The signal range of the brain alternating current signal is 0.03 Hz to 100 Hz. The 64-lead EEG signal amplifier specifically refers to a helmet-type dry electrode EEG cap with 64 dry electrodes.

7. The medical diagnosis auxiliary system based on EEG signal and artificial intelligence classification according to claim 6, characterized in that: The medical diagnosis assistance is used to perform medical diagnosis assistance based on the artificial intelligence classification results of the EEG signal, specifically to perform psychiatric medical diagnosis assistance based on the EEG signal classification data to obtain a diagnosis assistance report. The diagnosis assistance report is generated based on the EEG signal classification data, specifically referring to the psychiatric data features in the EEG of a single patient. The psychiatric data features include depressive components, anxiety components, schizophrenia components, affective disorder components and Alzheimer's disease components.