Alzheimer's classification system based on eeg data

Abstract

The application discloses an Alzheimer's disease classification system based on EEG data, comprising: a preprocessing module: the original EEG data is subjected to data standardization and denoising processing; a data cutting module: the preprocessed EEG data is subjected to non-repeated cutting according to the time length; a network architecture module: the network architecture module constitutes a deep learning model architecture, which comprises an input layer, a data processing layer and an output layer, and the data processing layer comprises a plurality of data processing modules connected in series in the front and back; a training module: used for inputting the cut EEG data into the network architecture module, training and adjusting the model parameters; a prediction module: used for inputting new EEG data into the trained model to perform AD, MCI and HA three-classification prediction. The application can more comprehensively and automatically extract the features of the EEG data, improve the sensitivity of the model to the time sequence characteristics through the preprocessing, cutting of the EEG data, and after multiple rounds of training, the model has high accuracy and stability in the three-classification task of Alzheimer's disease.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly to an Alzheimer's disease classification system based on EEG data. BACKGROUND

[0002] Dementia is a clinical syndrome characterized by a gradual decline in two or more cognitive domains, including memory, language, executive and visual-spatial functioning, personality and behavior, resulting in the loss of the ability to perform instrumental and / or basic activities of daily living. Dementia is caused by various diseases and injuries that affect the brain. Alzheimer's disease (AD) is the most common form of dementia. As the leading cause of dementia worldwide, the prevalence of Alzheimer's disease continues to grow to some extent due to global aging. The generally recognized intermediate stage between healthy aging (HA) and AD is called mild cognitive impairment (MCI). At present, it is not clear how many individuals with MCI will develop AD dementia. Multifaceted evidence suggests that modifying lifestyle at the early stage of the disease that increases risk and initiating drug and non-drug treatments, although cannot change its course, helps to maintain personal autonomy in daily activities and significantly reduces the overall cost of disease management. Alternative solutions are needed to explore other faster, cheaper, non-invasive biomarkers to achieve early detection of AD.

[0003] To achieve this goal, researchers have suggested analyzing electroencephalogram (EEG) signals to find features and biomarkers that can help diagnose AD. EEG signals are a non-invasive, time-varying, nonlinear, non-stationary electrical signal that can record brain electrical activity. More and more researchers have begun to use EEG to detect and distinguish dementia variants, especially the early detection of AD. However, due to the complexity, non-stationary and nonlinear characteristics of EEG signals, and accompanied by noise signals in the acquisition process, feature extraction of EEG signals becomes very difficult. The three main features between AD and HA in general EEG are: EEG deceleration, EEG complexity reduction and EEG synchrony. Traditional signal processing methods for detecting EEG deceleration, such as discrete Fourier transform (DFT), power spectral density (PSD) and continuous wavelet transform (CWT), etc. to extract features for detection. However, the above methods still have some shortcomings, DFT cannot provide local information about the signal because it only provides a frequency domain representation of the entire signal. PSD is usually based on the assumption of stationarity, that is, the statistical properties of the signal do not change over time. However, EEG signals are usually non-stationary, which can lead to limitations of PSD when analyzing non-stationary signals. The computational complexity of CWT is relatively high, especially for long time series. This can lead to infeasibility in real-time applications. When choosing wavelet scales, the balance between frequency and time needs to be considered. In recent years, with the rise of deep learning tasks, new solutions to the feature extraction task of EEG signals can be obtained. SUMMARY

[0004] The application aims to provide an Alzheimer's disease classification system based on EEG data, which effectively preprocesses, cuts, trains and classifies EEG data through a deep learning model, so as to improve the accuracy and stability of Alzheimer's disease classification detection.

[0005] The application achieves the above-mentioned purpose by the following technical solutions.

[0006] The Alzheimer's disease classification system based on EEG data comprises:

[0007] A preprocessing module: data standardization and denoising processing are performed on original EEG data.

[0008] A data cutting module: the preprocessed EEG data is cut without repetition according to time length to adapt to the input requirements of the network architecture module.

[0009] A network architecture module: the network architecture module constitutes a deep learning model architecture, which comprises an input layer, a data processing layer and an output layer, the data processing layer comprises a plurality of data processing modules connected in series from the beginning to the end, the data processing module comprises a Conv module and an Mlp module, the output of the Conv module is connected with the input of the Mlp module, and the linear layer of the output layer maps the high-dimensional features learned by the model to a category space, and then outputs a classification result after performing layer normalization.

[0010] A training module: used for inputting the cut EEG data into the network architecture module, adjusting the model parameters through multiple rounds of training, and improving the accuracy on the training set.

[0011] A prediction module: used for inputting new EEG data into the trained model to perform AD, MCI and HA three-classification prediction.

[0012] Further, the Conv module comprises a convolution layer for extracting input feature space information and a LayerNormalization layer for performing layer normalization.

[0013] Still further, the Mlp module comprises two fully connected layers, a StarReLU activation function and a dropout layer, the data passing through the previous fully connected layer is introduced into the StarReLU activation function to introduce a nonlinear transformation, and then input into the next fully connected layer, the output layer size of the next fully connected layer is consistent with the output size of the previous fully connected layer, and the two fully connected layers are mapped to each other.

[0014] Still further, the StarReLU activation function is:

[0015] StarReLU = s * (ReLU(x))2 +b

[0016] Wherein: s is a learnable scaling factor, and b is a bias term.

[0017] Further, the training module adopts an RMSprop optimizer to update the model parameters, and selects a cross-entropy loss function in model training to optimize the model.

[0018] Further, the RMSprop optimizer formula is as follows:

[0019]

[0020] Wherein, theta t is a parameter, gt is a gradient, eta is a learning rate, beta is a decay coefficient, epsilon is a smoothing term, mu t is momentum.

[0021] Further, the cross-entropy loss function is as follows:

[0022]

[0023] Wherein, is the predicted value of the model, y ij Data is the actual value, N is the sample quantity, and C is the category quantity

[0024] The present application has the following advantages: through a deep learning model, the features of EEG data can be more comprehensively and automatically extracted, the sensitivity of the model to time sequence features is improved through EEG data preprocessing and cutting, and after multiple rounds of training, the model has high accuracy and stability in the three-classification task of Alzheimer's disease. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 It is an algorithm schematic diagram of the network architecture module of the present application.

[0026] Figure 2 It is the training result of the model of the present application on a 2-second time length cut data set.

[0027] Figure 3 It is the test result of the model of the present application on a 2-second time length cut data set DETAILED DESCRIPTION

[0028] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations.

[0029] Therefore, the following detailed description of the embodiments of the application provided in the drawings is not intended to limit the scope of the application claimed, but merely represents selected embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the application.

[0030] It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0031] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0032] In the description of the application, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship commonly placed when the product of the application is used, or the orientation or positional relationship commonly understood by those skilled in the art, which is only for the convenience of describing the application and simplifying the description, and does not indicate or imply that the indicated device or element must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application. In addition, the terms "first", "second", etc. are only used for differentiation and cannot be understood as indicating or implying relative importance.

[0033] In the description of the application, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set", "mount", "connected", "connected" should be understood broadly, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication inside two elements. For those of ordinary skill in the art, the specific meaning of the above terms in the application can be understood according to the specific circumstances.

[0034] As Figure 1 shown, one embodiment of the application is as follows:

[0035] The Alzheimer's disease classification system based on EEG data includes:

[0036] The preprocessing module: the original EEG data is standardized by mean and standard deviation to eliminate the scale difference between different samples, and filtering or other signal processing techniques are used to reduce the influence of noise on the model.

[0037] Data cutting module: the pre-processed EEG data is cut according to the time length without repetition to adapt to the input requirements of the network architecture module, and the specific steps include: cutting the EEG data according to the specified time length, ensuring that the key information is not repeated, and optionally considering the sliding of the time window to cover more comprehensive electroencephalogram signals.

[0038] Network architecture module: the network architecture module constitutes a deep learning model architecture, which includes an input layer, a data processing layer and an output layer, the data processing layer includes a plurality of data processing modules connected in series from beginning to end, the data processing module includes a Conv module and an Mlp module, the output of the Conv module is connected with the input of the Mlp module, and the linear layer of the output layer maps the high-dimensional features learned by the model to the category space, performs layer normalization, and then outputs the classification result through the linear layer.

[0039] The Conv module includes a convolution layer for extracting spatial information of the input feature space and a LayerNormalization layer for performing layer normalization, and the Mlp module includes two fully connected layers, a StarReLU activation function and a dropout layer, the data after the previous fully connected layer is introduced into the non-linear transformation through the StarReLU activation function and input into the next fully connected layer, the output layer size of the next fully connected layer is consistent with the output size of the previous fully connected layer, and the two fully connected layers are mapped to each other.

[0040] The Conv module extracts the spatial information of the input features through convolution operation, the standardization and the activation function enhance the expression ability of the model, and the position transformation is used to introduce the position information of the sequence data, the Mlp module realizes the structure of the multi-layer perceptron, through this module, the network can perform non-linear transformation, learn the complex representation of the input data, and improve the fitting ability of the model.

[0041] As Figure 1As shown, the sheared electroencephalogram data is down-sampled by a Conv1 with 64 convolution kernels of size (7, 33). The down-sampled data is processed by layer normalization, which is applied to each time series sample to improve the training speed and accuracy of the model. The layer-normalized data is passed to an MLP layer, which is composed of two fully connected layers and a StarReLU activation function. When passing through the first fully connected layer, the number of neurons in the output layer is 4 times that of the input layer, and then the data is subjected to a StarReLU activation function to introduce a nonlinear transformation. The input layer of the second fully connected layer is the same size as the output layer of the first fully connected layer, and the output layer is the same as the first fully connected layer. The first fully connected layer and the second fully connected layer are mapped to each other. Next, the data is processed by a series of similar convolution blocks and MLP layers, each of which gradually increases the number of output channels to capture higher-level feature representations. The output layer uses two linear layers, where the output of the first linear layer is normalized and then converted to 3 dimensions by the final linear layer, corresponding to the outputs of the AD-MCI-NC three categories.

[0042] In this architecture, the output of the convolution block is connected to the input of the MLP layer, forming a series of module sequences, so that the model can effectively learn the multi-level representation of the input sequence and achieve deep processing and feature extraction of the sequence data.

[0043] The StarReLU activation function is:

[0044] StarReLU = s * (ReLU(x)) 2 +b

[0045] where s is a learnable scaling factor and b is a bias term, which enhances the model's ability to model non-linear features through the activation function.

[0046] Training module: used to input the sheared EEG data into the network architecture module, adjust the model parameters through multiple rounds of training, and improve the accuracy on the training set.

[0047] The training module uses the RMSprop optimizer to update the model parameters, and selects the cross-entropy loss function for model optimization during model training. The specific steps include: using the preprocessed and sheared EEG data for supervised training of the deep learning model, optimizing the model through appropriate loss functions such as cross-entropy loss, and updating the model parameters using the RMSprop optimizer.

[0048] The formula of the RMSprop optimizer is as follows:

[0049]

[0050] In the formula, θt is a parameter, gt is a gradient, η is a learning rate, β is a decay coefficient, ∈ is a smoothing term, and μ t is a momentum. Through the RMSprop optimizer, it adjusts the learning rate dynamically according to the historical information of the gradient to adapt to the change speed of different parameters, thereby more effectively updating the model parameters and improving the stability and convergence speed of training.

[0051] The cross-entropy loss function is:

[0052]

[0053] In the formula, is the predicted value of the model, y ij Data is the actual value, N is the sample number, and C is the number of categories. The cross-entropy loss is used to measure the difference between the model prediction and the actual label.

[0054] The prediction module is used to input new EEG data into the trained model to perform AD, MCI and HA three-classification prediction, and the specific steps include: adapting the format of the new EEG data through the data preprocessing and cutting module to the input of the network, using the trained deep learning model to perform Alzheimer's disease classification prediction, and outputting the prediction result, including the classification label and the probability probability.

[0055] The modular design makes the model have high flexibility and universality, and is suitable for different EEG data sets and classification tasks. Through the application, the diseases related to electroencephalogram can be more accurately diagnosed and classified, and important support is provided for early intervention and treatment.

[0056] The training and test results of the model of the application on the 2-second time length cut data set for the electroencephalogram AD-MCI-HA three-classification task are as follows:

[0057]

[0058] As can be seen from the above table, after multiple rounds of training, on the 2-second time length cut data set, the average training accuracy of the model of the application is 99.63%, and the average test accuracy is 97.71%, which has good stability and high accuracy.

[0059] Although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to part of the technical features, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

Claims

1. An Alzheimer's classification system based on EEG data, characterized by: The application relates to a deep learning model for AD, MCI and HA classification, which comprises the following steps: a preprocessing module: data standardization and denoising processing of original EEG data; a data cutting module: non-repeated cutting of the preprocessed EEG data according to the time length to adapt to the input requirements of the network architecture module; a network architecture module: the network architecture module constitutes a deep learning model architecture, which comprises an input layer, a data processing layer and an output layer, the data processing layer comprises a plurality of data processing modules connected in series, the data processing module comprises a Conv module and an Mlp module, the output of the Conv module is connected with the input of the Mlp module, the linear layer of the output layer maps the high-dimensional features learned by the model to a category space, and the classification result is output after layer normalization and linear layer output; the output layer adopts two linear layers, wherein the output of the first linear layer is converted into three dimensions through the final linear layer after normalization, corresponding to the outputs of the AD-MCI-NC three categories; the Conv module comprises a convolution layer for extracting input feature space information and a LayerNormalization layer for executing layer normalization, the Mlp module comprises two fully connected layers, a StarReLU activation function and a dropout layer, the data passing through the previous fully connected layer is introduced into the StarReLU activation function to introduce nonlinear transformation, and then input into the next fully connected layer, the output layer size of the next fully connected layer is consistent with the output size of the previous fully connected layer, and the two fully connected layers are mapped to each other; a training module: used for inputting the cut EEG data into the network architecture module, adjusting the model parameters through multiple rounds of training, and improving the accuracy on the training set; a prediction module: used for inputting new EEG data into the trained model to perform AD, MCI and HA three-classification prediction.

2. The Alzheimer's classification system based on EEG data according to claim 1, wherein: The StarReLU activation function is as follows: StarReLU = s * (ReLU(x)) 2 + b Wherein: s is a learnable scaling factor, and b is a bias term.

3. The Alzheimer's classification system based on EEG data according to claim 1, wherein: The training module adopts an RMSprop optimizer to update the model parameters, and a cross-entropy loss function is selected in the model training to optimize the model.

4. The Alzheimer's classification system based on EEG data according to claim 3, wherein: The RMSprop optimizer formula is as follows: where θtis the parameter, gtis the gradient, η is the learning rate, β is the decay coefficient, ∈ is the smoothing term, and μ t is the momentum.

5. The Alzheimer's classification system based on EEG data according to claim 3, wherein: The cross-entropy loss function is as follows: wherein is the predicted value of the model, y ij Data is the actual value, N is the number of samples, and C is the number of classes.

Images