A Tinnitus Classification Method and System Based on EEG Electroencephalography and Deep Learning

Abstract

This invention discloses a tinnitus grading method and system based on EEG electroencephalography (EEG) analysis and deep learning, relating to the field of tinnitus grading technology. The method includes: in a pre-training phase, converting EEG signals from tinnitus patients with different tinnitus severity levels into EEG images and inputting them into an EEG autoencoder to extract latent features from the EEG signals. These features are then input into a five-class linear classifier in a downstream fine-tuning phase to obtain the probability of each tinnitus severity level. The severity of tinnitus is predicted based on these probabilities. After training, a deep neural network model is obtained, comprising an EEG autoencoder for EEG feature extraction and a linear classifier for generating the severity of the patient's tinnitus. Using the deep neural network model, the probability of the tinnitus severity of the patient belonging to each of the five severity levels is obtained, thereby accurately predicting the severity of tinnitus and helping medical personnel develop personalized rehabilitation plans for patients.

Description

Technical Field

[0001] This invention relates to the field of tinnitus classification technology, and more specifically, to a tinnitus classification method and system based on EEG electroencephalogram analysis and deep learning. Background Technology

[0002] Tinnitus is a prevalent health problem affecting approximately 10% of the global population and is clinically considered a common but challenging symptom. Traditional tinnitus classification methods rely primarily on subjective patient scales, which are rather general and fail to accurately describe the heterogeneity of tinnitus, exhibiting issues such as high subjectivity and insufficient standardization. Although new classification advances have explored various aspects such as tinnitus perception, causal risk factors, tinnitus-related distress, and response to treatment, they have not yet fully elucidated the neural mechanisms of tinnitus. Therefore, a more accurate and sensitive method for objectively classifying tinnitus is needed.

[0003] EEG signals are non-invasive and provide real-time, objective reflections of brain activity, thus serving as an objective biomarker for tinnitus severity classification tasks. EEG maps constructed using EEG signals can simultaneously represent temporal and spatial information, extracting high-level features that are difficult to capture using traditional signal processing methods, aiding in the analysis of dynamic changes in brain activity and network structure. Generative Neural Networks (GCNs) are extensions of convolutional neural networks combined with spectral theory. Using GCNs, discriminative features from EEG maps can be extracted from both the temporal and spatial domains. Therefore, how to utilize the GCN framework for EEG signal feature extraction to achieve accurate tinnitus classification is a problem that needs to be solved. Summary of the Invention

[0004] To address the aforementioned technical issues, this invention proposes a tinnitus grading method and system based on EEG electroencephalography and deep learning, which enables accurate prediction of the severity of tinnitus in patients and helps medical personnel develop personalized rehabilitation plans for patients.

[0005] The first aspect of this invention provides a tinnitus classification method based on EEG electroencephalogram analysis and deep learning, comprising the following steps:

[0006] We collected EEG signals from tinnitus patients with different tinnitus severity at rest and preprocessed them, then converted the preprocessed EEG signals into EEG images.

[0007] In the pre-training stage, the EEG image is input into an EEG autoencoder based on the GCN framework for pre-training to obtain an EEG autoencoder for EEG feature extraction. In the downstream fine-tuning stage, a five-class linear classifier is trained to obtain a linear classifier for generating the severity of the patient's tinnitus. A deep neural network model is constructed using the trained EEG autoencoder and the linear classifier.

[0008] The preprocessed EEG signals of patients with tinnitus to be graded are obtained, converted into EEG images, and input into a deep neural network model. The model outputs the probability that the severity of tinnitus in the patients belongs to one of the five severity levels, thus predicting the severity of tinnitus.

[0009] In this protocol, EEG signals from tinnitus patients with different degrees of tinnitus at rest were collected and preprocessed, specifically as follows:

[0010] We collected resting EEG signals from tinnitus patients with different tinnitus severity. We then cleaned the collected EEG signals to remove redundant and erroneous data and eliminated noise and interference from the EEG signals through mean filtering.

[0011] The filtered EEG signal is segmented, and the segmented EEG signal is baseline corrected to eliminate baseline offset. Then the data is re-referenced to align the EEG signal in the standard coordinate system. During the processing, interpolation methods are used to estimate and replace missing data.

[0012] Independent component analysis is used to extract independent components from the aligned EEG signal. Damaged or noisy independent components are then removed from the independent components to obtain and process the EEG signal.

[0013] In this scheme, the preprocessed EEG signal is converted into an EEG image, specifically as follows:

[0014] Obtain pre-processed EEG signals X∈R from tinnitus patients with different tinnitus severity. C×T Where C is the number of electrodes and T is the number of sampling points for each electrode;

[0015] The preprocessed EEG signal is represented by a graph, with each electrode regarded as a node, and an edge structure is constructed based on the correlation between the electrodes. The resulting EEG graph is denoted as G = (X, A), where X represents the attribute of the node and A represents the weighted adjacency matrix of the edge structure.

[0016] For the weighted adjacency matrix A, where matrix element a ij Represents node X i and node X j The hysteresis coherence between EEG signals is expressed as:

[0017] a ij (τ)=E[X i (t)·X j (t+τ)]

[0018] Where X i (t) represents node X i The EEG signal at time t, Xj (t+τ) represents X j For a signal delayed by τ time units, E[·] represents the expectation operator of the average value of the random variable.

[0019] In this scheme, during the pre-training phase, the EEG image is input into an EEG autoencoder based on the GCN framework for pre-training to obtain an EEG autoencoder for EEG feature extraction, specifically as follows:

[0020] An encoder and decoder for an EEG autoencoder are constructed based on the GCN framework. The encoder consists of two GCN layers. During the pre-training phase, the encoder is used to encode the EEG image to obtain a low-dimensional feature representation of the EEG image, represented as follows:

[0021]

[0022] Where H (l) This represents the input feature matrix of the l-th layer. Add the adjacency matrix of the self-loop I. express The corresponding degree matrix, Let σ represent the learnable weight matrix, and σ represent the ReLU activation function.

[0023] The decoder includes an attribute decoder and a structure decoder. The attribute decoder is used to reconstruct node attributes from the encoded low-dimensional feature representation. A series of deconvolutions are applied to expand the data back to the original feature space, as shown below:

[0024]

[0025] in, Represents the reconstructed node feature matrix;

[0026] The structure decoder is used to reconstruct the adjacency matrix based on the relationships between nodes in the EEG graph obtained from the encoded low-dimensional feature representation, thereby restoring the connectivity relationships in the EEG graph. The structure decoder predicts whether there is connectivity between each pair of nodes, expressed as:

[0027]

[0028] Among them, z i ,z j Represents node X i and node X j The latent feature representation, where P represents the probability that there is a connection between the nodes, Represents node X after reconstruction i and node X j The connectivity relationship;

[0029] In the decoder of the EEG autoencoder, the prediction layer is trained based on the output of the attribute decoder Z, as follows:

[0030]

[0031] in, This represents the reconstructed adjacency matrix.

[0032] In this scheme, the original node feature matrix X and the reconstructed node feature matrix are used in the EEG autoencoder. The reconstruction validity of the L2 norm balanced attributes between them, where the reconstruction validity R of the attributes is... A Represented as:

[0033]

[0034] Using the original adjacency matrix A and the reconstructed adjacency matrix The reconstruction effectiveness of the L2 norm equilibrium structure between them, where the reconstruction effectiveness R of the structure is... S Represented as:

[0035]

[0036] The reconstruction is comprehensively evaluated from both attribute and structural perspectives. A loss function for the EEG autoencoder is defined, and the EEG autoencoder is iteratively trained until the loss function converges. The loss function L is expressed as:

[0037] L=αR A +(1-α)R S

[0038] Where α represents a hyperparameter, used to balance the importance of attribute reconstruction and structural reconstruction.

[0039] In this scheme, during the downstream fine-tuning stage, a five-class linear classifier is trained to obtain a linear classifier used to generate the severity of the patient's tinnitus, specifically as follows:

[0040] Freeze the encoder portion of the pre-trained EEG autoencoder, use the encoder portion to generate latent feature representations of the EEG graph, calculate the average value of the latent feature representations, and obtain an m-dimensional vector. The m-dimensional vector is input into a three-layer multilayer perceptron to train a five-class linear classifier;

[0041] The structure of the linear classifier is as follows: y (l) =σ(W (l) x (l) +b (l) ), where y (l) W represents the output of the l-th layer. (l)Let x represent the weight matrix of the l-th layer. (l) Let b represent the m-dimensional input vector of the l-th layer. (l) Let σ represent the bias vector of the l-th layer, and let σ represent the activation function.

[0042] The probability of each tinnitus severity level output by the classifier is obtained through iterative training, and the classification performance of the linear classifier is verified using validation data. When the classification performance meets the preset standard, a linear classifier is obtained to generate the severity of the patient's tinnitus.

[0043] In this scheme, a deep neural network model is constructed based on a pre-trained encoder and a linear classifier;

[0044] Acquire preprocessed EEG signals from patients with tinnitus to be graded, and convert the EEG signals into EEG images as input to a deep neural network model;

[0045] The deep neural network model is used to calculate and output the probability that the tinnitus severity of the patient to be classified belongs to one of the five severity levels, thereby obtaining the tinnitus severity prediction result of the patient to be classified.

[0046] The second aspect of this invention provides a tinnitus grading system based on EEG electroencephalogram analysis and deep learning. The system includes a data acquisition unit, an image processing unit, a tinnitus grading unit, and an output unit.

[0047] The data acquisition unit is responsible for acquiring the EEG signals of patients with tinnitus to be graded and preprocessing the acquired EEG signals.

[0048] The graph processing unit is responsible for converting the preprocessed EEG signal into an EEG graph.

[0049] The tinnitus grading unit is responsible for constructing an EEG autoencoder based on the GCN framework, training the EEG autoencoder through pre-training, training a five-class linear classifier in the downstream fine-tuning stage, constructing a deep neural network model using the trained EEG autoencoder and linear classifier, and calculating the probability that the tinnitus severity of the tinnitus patient to be graded belongs to one of the five severity levels using the trained deep neural network model.

[0050] The output unit is responsible for determining the severity of tinnitus based on the probability that the tinnitus patient belongs to one of the five severity levels, and outputting the tinnitus grading result.

[0051] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0052] This invention generates a high-quality, large-scale EEG dataset from resting-state EEG data of tinnitus patients with varying degrees of tinnitus severity. In the pre-training phase, the converted EEG images are input into an EEG autoencoder to extract latent features from the EEG data and generate simplified yet accurate representations. These features are then fed into a five-class linear classifier in the downstream fine-tuning phase to obtain the probability of each tinnitus severity level, predicting the severity of tinnitus based on these probabilities. After training, a deep neural network model is obtained, comprising an EEG autoencoder for EEG feature extraction and a linear classifier for generating the severity of the patient's tinnitus. When a user inputs the preprocessed EEG signal from a tinnitus patient, the EEG signal is first converted into an EEG image, which is then used as input to the aforementioned deep neural network model. Finally, the model outputs the probability that the patient's tinnitus severity belongs to each of the five severity levels, thereby predicting the severity of tinnitus and helping medical personnel develop personalized rehabilitation plans for patients. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0054] Figure 1 A flowchart of a tinnitus grading method based on EEG electroencephalography and deep learning is shown;

[0055] Figure 2 A flowchart illustrating the tinnitus severity grading using EEG signals is shown.

[0056] Figure 3 A schematic diagram of the reconstruction loss of the EEG autoencoder is shown;

[0057] Figure 4 A flowchart of a tinnitus classification system based on EEG electroencephalography and deep learning is shown. Detailed Implementation

[0058] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0059] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0060] like Figure 1 As shown, the first embodiment of the present invention provides a tinnitus grading method based on EEG electroencephalogram analysis and deep learning, including:

[0061] S102, Collect resting EEG signals from tinnitus patients with different tinnitus severity and preprocess them, then convert the preprocessed EEG signals into EEG images;

[0062] S104, In the pre-training stage, the EEG image is input into the EEG autoencoder based on the GCN framework for pre-training to obtain the EEG autoencoder for EEG feature extraction. In the downstream fine-tuning stage, a five-class linear classifier is trained to obtain a linear classifier for generating the severity of the patient's tinnitus. The trained EEG autoencoder and linear classifier are used to build a deep neural network model.

[0063] S106: Obtain the preprocessed EEG signal of the tinnitus patient to be graded, convert it into an EEG image and input it into a deep neural network model, output the probability that the severity of the tinnitus of the patient to be graded belongs to one of the five severity levels, and predict the severity of the tinnitus.

[0064] It should be noted that resting-state EEG signals from tinnitus patients with varying degrees of tinnitus severity were collected. The collected EEG signals underwent data cleaning to remove redundant and erroneous data. Mean filtering was used to eliminate noise and interference in the EEG signals, thereby extracting useful signal features. The filtered EEG signals were then segmented, with a preferred sampling frequency of 1278Hz. Each segment contained two seconds of data, generating 256 power values ​​per segment. Baseline correction was performed on the segmented EEG signals to eliminate baseline shift and enhance data reliability. The data was then re-referenced to align the EEG signals within a standard coordinate system for easier comparison and integration. Interpolation methods were used to estimate and replace missing data during processing. Independent component analysis was used to extract independent components from the aligned EEG signals to further reveal the underlying structure. Damaged or noisy independent components were removed from these components, ensuring the purity of the dataset.

[0065] It should be noted that the pre-processed EEG signals X∈R of tinnitus patients with different degrees of tinnitus severity were obtained. C×T , where R C×T Let the space be a real number space of size C×T, where C is the number of electrodes and T is the number of sampling points for each electrode. The preprocessed EEG signal is represented as a graph, with each electrode as a node, and an edge structure is constructed based on the correlation between electrodes. The resulting EEG graph is denoted as G=(X,A), where X represents the attribute of the node and A represents the weighted adjacency matrix of the edge structure.

[0066] Because the number of original sampling time points is too large, the raw data cannot be directly used as the temporal feature of each node. Therefore, data compression is performed, the average value of the data within each segment is calculated, and this average value is used as one dimension of the node feature vector. Data compression provides temporal features describing EEG fluctuations over a longer period, while omitting short-term detailed changes. For the weighted adjacency matrix A, the matrix element a... ij Represents node X i and node X j The hysteresis coherence between EEG signals is expressed as:

[0067] a ij (τ)=E[X i (t)·X j (t+τ)]

[0068] Where X i (t) represents node X i The EEG signal at time t, X j (t+τ) represents X j For a signal delayed by τ time units, E[·] represents the expectation operator of the average value of the random variable.

[0069] like Figure 2 As shown, the EEG autoencoder consists of an encoder and a decoder, both built on the GCN framework. During the pre-training phase, the EEG image is input into the GCN-based EEG autoencoder for pre-training to obtain the EEG autoencoder used for EEG feature extraction, specifically:

[0070] An encoder and decoder for an EEG autoencoder are constructed based on the GCN framework. The encoder consists of two GCN layers, which effectively model the attributes of individual nodes and their interactions. The application of GCN in the encoder effectively reduces the dimensionality of the EEG data while preserving basic features. During the pre-training phase, the encoder is used to encode the EEG graph, obtaining a low-dimensional feature representation of the EEG graph, expressed as:

[0071]

[0072] Where H (l) This represents the input feature matrix of the l-th layer. Add the adjacency matrix of the self-loop I. express The corresponding degree matrix, Let σ represent the learnable weight matrix, and σ represent the ReLU activation function.

[0073] The decoder decodes the low-dimensional feature representation, attempting to restore the data to a state as close as possible to the pre-encoded data. If the restored data is very similar to the pre-encoded data, it can be considered that the encoder extracted the key information of the original data well, and the decoder has good recovery capabilities. The decoder includes an attribute decoder and a structure decoder. The attribute decoder is used to reconstruct node attributes from the encoded low-dimensional feature representation, and a series of deconvolutional layers are applied to expand the data back to the original feature space, as shown below:

[0074]

[0075] in, Represents the reconstructed node feature matrix;

[0076] The structure decoder is used to reconstruct the adjacency matrix based on the relationships between nodes in the EEG graph obtained from the encoded low-dimensional feature representation, thereby restoring the connectivity relationships in the EEG graph. The structure decoder predicts whether there is connectivity between each pair of nodes, expressed as:

[0077]

[0078] Among them, z i ,z j Represents node X i and node X j The latent feature representation, where P represents the probability that there is a connection between the nodes, Represents node X after reconstruction i and node X j The connectivity relationship;

[0079] In the decoder of the EEG autoencoder, the prediction layer is trained based on the output of the attribute decoder Z, as follows:

[0080]

[0081] in, This represents the reconstructed adjacency matrix.

[0082] like Figure 3 As shown, in the EEG autoencoder, the original node feature matrix X and the reconstructed node feature matrix are used to... The reconstruction validity of the L2 norm balanced attributes between them, where the reconstruction validity R of the attributes is... A Represented as:

[0083]

[0084] Using the original adjacency matrix A and the reconstructed adjacency matrix The reconstruction effectiveness of the L2 norm equilibrium structure between them, where the reconstruction effectiveness R of the structure is...S Represented as:

[0085]

[0086] The reconstruction is comprehensively evaluated from both attribute and structural perspectives. A loss function for the EEG autoencoder is defined, and the EEG autoencoder is iteratively trained until the loss function converges. The loss function L is expressed as:

[0087] L=αR A +(1-α)R S

[0088] Here, α represents a hyperparameter, which is used to balance the importance of attribute reconstruction and structural reconstruction when evaluating the overall reconstruction quality.

[0089] In the downstream fine-tuning stage, the encoder portion of the pre-trained EEG autoencoder is frozen, and the encoder portion is used to generate the latent feature representation X∈R of the EEG graph. C×M , where R C×M Let m represent a real space of size C × M, where C is the number of nodes and M is the dimension of the encoded features. To classify the latent feature representations, the average value of the latent feature representations is calculated along the first dimension to obtain an m-dimensional vector. X i This represents the latent feature representation of the i-th node.

[0090] The m-dimensional vector is input into a three-layer multilayer perceptron to train a five-class linear classifier. The structure of the linear classifier is: y (l) =σ(W (l) x (l) +b (l) ), where y (l) W represents the output of the l-th layer. (l) Let x represent the weight matrix of the l-th layer. (l) Let b represent the m-dimensional input vector of the l-th layer. (l) Let represent the bias vector of the l-th layer, and σ represent the ReLU activation function. The probability of each tinnitus severity level output by the classifier is obtained through iterative training. The tinnitus severity level with the highest probability is selected as the final diagnosis result. The classification performance of the linear classifier is verified using validation data. When the classification performance meets the preset standard, a linear classifier used to generate the severity of the patient's tinnitus is obtained.

[0091] According to an embodiment of the present invention, the preprocessed EEG signal of a tinnitus patient to be graded is obtained, and the EEG signal is converted into an EEG image as input to a deep neural network model; the deep neural network model is used to calculate and output the probability that the tinnitus severity of the tinnitus patient to be graded belongs to one of the five severity levels, which is used as a severity score, and the severity category corresponding to the maximum severity score is obtained as the tinnitus severity prediction result of the tinnitus patient to be graded.

[0092] Figure 4 A flowchart of a tinnitus classification system based on EEG electroencephalography and deep learning is shown.

[0093] The second embodiment of the present invention provides a tinnitus grading system 4 based on EEG electroencephalogram analysis and deep learning. The system includes a data acquisition unit 401, an image processing unit 402, a tinnitus grading unit 403 and an output unit 404.

[0094] The data acquisition unit is responsible for acquiring the EEG signals of patients with tinnitus to be graded and preprocessing the acquired EEG signals.

[0095] The graph processing unit is responsible for converting the preprocessed EEG signal into an EEG graph.

[0096] The tinnitus grading unit is responsible for constructing an EEG autoencoder based on the GCN framework, training the EEG autoencoder through pre-training, training a five-class linear classifier in the downstream fine-tuning stage, constructing a deep neural network model using the trained EEG autoencoder and linear classifier, and calculating the probability that the tinnitus severity of the tinnitus patient to be graded belongs to one of the five severity levels using the trained deep neural network model.

[0097] The output unit is responsible for determining the severity of tinnitus based on the probability that the tinnitus patient belongs to one of the five severity levels, and outputting the tinnitus grading result.

[0098] The third embodiment of the present invention provides a computer-readable storage medium, which includes a tinnitus classification method program based on EEG electroencephalogram analysis and deep learning. When the tinnitus classification method program based on EEG electroencephalogram analysis and deep learning is executed by a processor, it implements the steps of the tinnitus classification method based on EEG electroencephalogram analysis and deep learning.

[0099] In the several embodiments provided in this application, it should be understood that the disclosed methods can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, and may be electrical, mechanical, or other forms.

[0100] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] Alternatively, if the integrated units of the present invention are implemented as software functional units and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0102] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A tinnitus grading method based on EEG electroencephalogram analysis and deep learning, characterized in that, Includes the following steps: We collected EEG signals from tinnitus patients with different tinnitus severity at rest and preprocessed them, then converted the preprocessed EEG signals into EEG images. Specifically, the preprocessed EEG signal is converted into an EEG image, which is done as follows: Obtain pre-processed EEG signals from tinnitus patients with different tinnitus severity. ,in The number of electrodes, The number of sampling points for each electrode; The preprocessed EEG signal is represented graphically, with each electrode considered as a node, and an edge structure is constructed based on the correlation between the electrodes. The resulting EEG graph is denoted as . ,in Represents the attributes of a node. The weighted adjacency matrix represents the edge structure; For the weighted adjacency matrix , of which matrix elements Represents a node and nodes The hysteresis coherence between EEG signals is expressed as: ， in Represents a node In time EEG signal at that time express In delay The signal after one time unit The expectation operator represents the mean of a random variable; During the pre-training phase, the EEG image is input into an EEG autoencoder based on the GCN framework for pre-training to obtain an EEG autoencoder for EEG feature extraction. The EEG autoencoder includes an encoder and a decoder. The encoder consists of two GCN layers. During the pre-training phase, the encoder is used to encode the EEG image to obtain a low-dimensional feature representation of the EEG image. ; The decoder includes an attribute decoder and a structure decoder. The attribute decoder is used to reconstruct node attributes from the encoded low-dimensional feature representation, and a series of deconvolutions are applied to expand the data back to the original feature space. The structure decoder is used to obtain the relationship between nodes in the EEG graph based on the encoded low-dimensional feature representation to reconstruct the adjacency matrix, restore the connectivity in the EEG graph, and predict whether there is a connection between each pair of nodes. In the EEG autoencoder, the original node feature matrix is ​​used... With reconstructed node feature matrix The reconstruction validity of the L2 norm balanced attributes, where the reconstruction validity of the attributes... Represented as: ， Through the original adjacency matrix Reconstructing the adjacency matrix The reconstruction effectiveness of the L2 norm equilibrium structure between them, where the reconstruction effectiveness of the structure Represented as: ， The reconstruction is comprehensively evaluated from both attribute and structural perspectives. A loss function for the EEG autoencoder is defined, and the EEG autoencoder is iteratively trained until the loss function converges. Represented as: ， in This represents hyperparameters, used to balance the importance of attribute reconstruction and structural reconstruction; In the downstream fine-tuning stage, a five-class linear classifier is trained to obtain a linear classifier used to generate the severity of the patient's tinnitus. A deep neural network model is then built using the trained EEG autoencoder and the linear classifier. The preprocessed EEG signals of patients with tinnitus to be graded are obtained, converted into EEG images, and input into a deep neural network model. The model outputs the probability that the severity of tinnitus in the patients belongs to one of the five severity levels, thus predicting the severity of tinnitus.

2. The tinnitus classification method based on EEG electroencephalogram analysis and deep learning according to claim 1, characterized in that, Resting-state EEG signals were collected from tinnitus patients of varying tinnitus severity and preprocessed, specifically as follows: We collected resting EEG signals from tinnitus patients with different tinnitus severity. We then cleaned the collected EEG signals to remove redundant and erroneous data and eliminated noise and interference from the EEG signals through mean filtering. The filtered EEG signal is segmented, and the segmented EEG signal is baseline corrected to eliminate baseline offset. Then the data is re-referenced to align the EEG signal in the standard coordinate system. During the processing, interpolation methods are used to estimate and replace missing data. Independent components are extracted from the aligned EEG signal using independent component analysis. Damaged or noisy independent components are then removed from the independent components to obtain a preprocessed EEG signal.

3. The tinnitus classification method based on EEG electroencephalogram analysis and deep learning according to claim 1, characterized in that, In the downstream fine-tuning stage, a five-class linear classifier is trained to obtain a linear classifier used to generate the severity of the patient's tinnitus, specifically: Freeze the encoder portion of the pre-trained EEG autoencoder, use the encoder portion to generate latent feature representations of the EEG graph, calculate the average value of the latent feature representations, and obtain an m-dimensional vector. The m-dimensional vector is input into a three-layer multilayer perceptron to train a five-class linear classifier; The structure of the linear classifier is as follows: ,in Indicates the first The output of the layer, Indicates the first The weight matrix of the layer, Indicates the first The m-dimensional input vector of the layer Indicates the first The layer's bias vector, Indicates the activation function; The probability of each tinnitus severity level output by the classifier is obtained through iterative training, and the classification performance of the linear classifier is verified using validation data. When the classification performance meets the preset standard, a linear classifier is obtained to generate the severity of the patient's tinnitus.

4. The tinnitus classification method based on EEG electroencephalogram analysis and deep learning according to claim 1, characterized in that, Construct a deep neural network model based on a pre-trained encoder and linear classifier; Acquire preprocessed EEG signals from patients with tinnitus to be graded, and convert the EEG signals into EEG images as input to a deep neural network model; The deep neural network model is used to calculate and output the probability that the tinnitus severity of the patient to be classified belongs to one of the five severity levels, thereby obtaining the tinnitus severity prediction result of the patient to be classified.

5. A tinnitus grading system based on EEG electroencephalogram analysis and deep learning, characterized in that, To implement the tinnitus classification method based on EEG electroencephalography and deep learning as described in any one of claims 1-4, the system includes a data acquisition unit, an image processing unit, a tinnitus classification unit, and an output unit. The data acquisition unit is responsible for acquiring the EEG signals of patients with tinnitus to be graded and preprocessing the acquired EEG signals. The graph processing unit is responsible for converting the preprocessed EEG signal into an EEG graph. The tinnitus grading unit is responsible for constructing an EEG autoencoder based on the GCN framework, training the EEG autoencoder through pre-training, training a five-class linear classifier in the downstream fine-tuning stage, constructing a deep neural network model using the trained EEG autoencoder and linear classifier, and calculating the probability that the tinnitus severity of the tinnitus patient to be graded belongs to one of the five severity levels using the trained deep neural network model. The output unit is responsible for determining the severity of tinnitus based on the probability that the tinnitus patient belongs to one of the five severity levels, and outputting the tinnitus grading result.

Images