Method for identifying autism spectrum disorder based on time sequence

Abstract

The invention belongs to the field of auxiliary diagnosis of autism spectrum disorders, and particularly relates to a time sequence-based autism spectrum disorder recognition method, which comprises the following steps of: constructing and training an autism recognition model, and inputting to-be-recognized data into the trained autism recognition model to obtain a recognition result; the autism identification model comprises a graph attention space network feature extractor, a bidirectional long and short term network time feature extractor and a multi-layer perceptron network classifier; according to the method, potential graph structure information and time sequence dependence among all the regions of the brain region template can be comprehensively utilized, and the method has the advantages in the aspect of capturing important features in time sequence data and complex relations among nodes.

Description

Technical Field

[0001] The present invention belongs to the field of auxiliary diagnosis of autism spectrum disorders, and specifically relates to a method for identifying autism spectrum disorders based on time series. Background Art

[0002] Autism spectrum disorder (ASD) is a serious neurological disease. The latest large-scale survey shows that the prevalence of autism spectrum disorder is about 1%-2%. Its clinical manifestations mainly include difficulties in social interaction, verbal communication, and repetitive behaviors. Due to the overlap of symptoms and the lack of clear biological tests that can be used as clear measurement criteria, the diagnosis of autism spectrum disorder is mainly performed by professionals using tools such as autism diagnostic observation and autism diagnostic interview. First, a large number of high-quality medical resources are indispensable. Secondly, this test lacks objectivity. Therefore, computer-aided diagnosis (CAD) technology is urgently needed to assist medical staff in auxiliary diagnosis. Neuroimaging technology provides a powerful technical means for understanding brain mechanisms. Resting-state functional magnetic resonance imaging (fMRI) is a commonly used non-invasive neuroimaging technology that provides better spatial resolution. It can observe functional changes in brain activity during task execution by measuring changes in blood oxygen level dependent (BOLD) contrast images.

[0003] There are two main research routes in the study of ASD auxiliary diagnosis based on fMRI: machine learning and deep learning. Due to the lack of data and the limitation of machine performance, early studies used traditional machine learning techniques on FC data from a single international neuroimaging laboratory, including supervised methods such as SVM, RF, logistic regression, and unsupervised methods such as PCA. In recent years, with the emergence of the ABIDE dataset and the widespread adoption of end-to-end deep learning applications, ASD diagnostic research has undergone a paradigm shift. Most deep learning methods have been applied to the task of identifying ASD patients using fMRI data. These methods include DNN, CNN, LSTM, etc. In recent years, with the rise of Transformer and GCN, their corresponding applications have also been used to identify ASD patients. Usually, the region of interest (ROI) of fMRI data is partitioned using different brain partitioning templates to obtain different ROI average time series. However, existing methods do not fully consider the potential spatial relationship between different ROIs, thereby ignoring the possible spatial characteristics of the time series data of fMRI data.

[0004] In order to solve these problems, the present invention proposes a time series-based graph attention network ASD autism spectrum disorder identification method, which can effectively extract and capture the spatial-temporal features in the time series data after extracting the time series fMRI data according to the brain partition template. Summary of the invention

[0005] To solve the above problems, the present invention provides a method for identifying autism spectrum disorders based on time series, including constructing and training an autism recognition model, inputting the data to be recognized into the trained autism recognition model to obtain a recognition result; the autism recognition model includes a graph attention spatial network feature extractor, a bidirectional long-short term network temporal feature extractor and a multi-layer perceptron network classifier;

[0006] The training process of the autism recognition model includes:

[0007] S1. Acquire fMRI data and preprocess it to obtain preprocessed data;

[0008] S2. Divide the whole brain into 200 brain regions according to the CC200 brain partition template, and extract the average time series data of each brain region based on the preprocessed data;

[0009] S3. Calculate the Pearson correlation coefficient of each brain region based on the average time series data, and construct a graph structure based on the Pearson correlation coefficient;

[0010] S4. Input the graph structure into the graph attention spatial network feature extractor to update the node features of each node;

[0011] S5. Input the node features of the graph structure into the bidirectional long-short term network time feature extractor to obtain a global context vector;

[0012] S6. Input the global context vector into the multi-layer perceptron network classifier to obtain the recognition result;

[0013] S7. Calculate the loss based on the recognition results and train the model parameters until the model parameters converge.

[0014] Beneficial effects of the present invention:

[0015] The present invention captures the functional connectivity between different brain regions through fMRI data, uses graph attention network and bidirectional LSTM network to obtain the spatial and temporal relationship between different brain regions, realizes the information enhancement between regions under the brain region template partitioning, thereby improving the accuracy of the model in ASD autism spectrum disorder identification based on fMRI time series data. In addition, the present invention optimizes the feature extraction ability of the model through the attention mechanism, and increases the generalization ability of the model during classification by combining dropout and the like. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flow chart of the method of the present invention;

[0017] Figure 2 It is a schematic diagram of the model structure of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The present invention provides an autism spectrum disorder identification method based on time series, including constructing and training an autism identification model, inputting the data to be identified into the trained autism identification model to obtain the identification result; the autism identification model includes a graph attention spatial network feature extractor, a bidirectional long-short term network temporal feature extractor and a multi-layer perceptron network classifier.

[0020] like Figure 1 , Figure 2 As shown, the training process of the autism recognition model includes:

[0021] S1. Acquire fMRI data and preprocess it to obtain preprocessed data.

[0022] Specifically, the present invention uses the ABIDE I dataset as experimental data. The ABIDE I dataset comes from 17 different international brain imaging laboratories and contains a total of 1,112 subjects, of which 539 are ASD patients and 573 are normal subjects.

[0023] The fMRI data of each subject was preprocessed, including:

[0024] Delete the brain slice data of the first 10 time points in the fMRI data to avoid the influence of the subjects' inability to enter the resting state immediately at the beginning of the test;

[0025] The brain slice data obtained at different time points are time-corrected in the form of interlaced scanning; all brain slice data are arranged in chronological order, and the brain slice data in the middle of the order is generally selected as a reference when performing interlaced scanning;

[0026] To prevent the fMRI data from being disturbed by the subject's head movement, a rigid transformation was used to fix the brain in all images at the same target position. Then, the MNI image was used as a configuration template to relocate the structural and functional images and remove the scalp.

[0027] Noise regression and spatial normalization were used to map the brains of different subjects into the same space. A 6mm×6mm×6mm filter was selected to further remove artifacts. Low-frequency signals in the range of 0.01-0.08Hz were then selected for filtering to remove the influence of high-frequency noise such as breathing and heartbeat.

[0028] S2. The whole brain was divided into 200 brain regions according to the CC200 brain partition template, and the average time series data of each brain region was extracted based on the preprocessed data.

[0029] S3. Calculate the Pearson correlation coefficient of each brain region based on the average time series data, and construct the graph structure based on the Pearson correlation coefficient.

[0030] Specifically, the specific process of constructing the graph structure in step S3 includes:

[0031] Each brain region is taken as a node, and the average time series data is used as the node feature; the Pearson correlation coefficient between each two nodes is calculated based on the time series data. If the Pearson correlation coefficient is greater than 0.5, an edge is connected between the two nodes, and the Pearson correlation coefficient is recorded as the edge value; if the Pearson correlation coefficient is not greater than 0.5, there is no connecting edge between the two nodes; thus, the adjacency matrix of the graph structure is obtained.

[0032] S4. Input the graph structure into the graph attention spatial network feature extractor to update the node features of each node.

[0033] Specifically, the adjacency matrix is ​​normalized to obtain a normalized matrix, which is expressed as

[0034]

[0035] Where A represents the adjacency matrix, represents the normalized adjacency matrix, and D represents the degree matrix of the graph structure.

[0036] Specifically, there are potential spatial connections between different brain regions. The graph attention spatial network spatial feature extractor not only aggregates the potential information of each brain region, but also determines the influence of different brain regions on ASD identification through the attention mechanism, and further extracts effective spatial features. The graph attention spatial network feature extractor includes multiple attention graph convolution layers. Step S4 specifically includes

[0037] For each node, its node feature is used as the initial input, and after passing through multiple attention graph convolution layers, the output of the last attention graph convolution layer is used as the updated node feature. The processing process of the lth attention graph convolution layer includes:

[0038] S31. Output of node i in the l-1th attention graph convolution layer As the input of the lth attention graph convolution layer;

[0039] S32. Input of node i in the lth attention graph convolution layer The attention mechanism is used to perform neighbor node feature fusion processing to obtain the output of node i in the lth attention graph convolution layer. Expressed as

[0040]

[0041] in, represents the input of node i in the lth attention graph convolution layer, W (l) represents the trainable parameters, represents the linear transformation of node i in the lth attention graph convolution layer, [||] represents the concatenation process,

[0042] represents the weight vector of the lth attention graph convolutional layer, LeakyReLU is the activation function, N(i) represents the set of neighbor nodes of node i, represents the attention weight between node i and node j, represents the updated feature representation of node i in the lth attention graph convolutional layer.

[0043] S5. Input the node features of the graph structure into the bidirectional long-term and short-term network temporal feature extractor to obtain the global context vector.

[0044] Specifically, step S5 includes:

[0045] S51. All node features are passed through a bidirectional LSTM network to obtain a comprehensive temporal representation at different time steps;

[0046] Specifically, each node feature processed by the graph attention spatial network feature extractor contains spatial features at different time steps. The spatial features corresponding to different time steps are extracted from all node features to form a sequence of spatial features at different time steps. Assume that the node feature after the update of node i is X i ={x 1i ,x 2i ,…,x Ti}, x ti represents the spatial feature of node i at time step t, T represents the number of time steps; then the spatial feature sequence of time step t is expressed as H t =(x t1 ,x t2 ,…,x tM ), M represents the number of nodes. The spatial feature sequences at different time steps are passed through a bidirectional LSTM network to obtain the temporal comprehensive representation at different time steps.

[0047] S52. Calculate the attention score of the temporal comprehensive representation of each time step, expressed as

[0048] e t =tanh(W a h t +b a )

[0049] Among them, e t represents the attention score at the tth time step, W a represents the learnable weight matrix, b a represents the learnable bias vector, h t represents the time-integrated representation of the t-th time step;

[0050] S53. Normalize each attention score to get a scalar score, expressed as

[0051]

[0052] Among them, α t represents the scalar score of the tth time step, T represents the number of time steps;

[0053] S54. A global context vector is obtained by weighted calculation based on the scalar score and the temporal comprehensive representation, which is expressed as

[0054]

[0055] Where c represents the global context vector.

[0056] S6. Input the global context vector into the multi-layer perceptron network classifier to obtain the recognition result.

[0057] Specifically, the multilayer perceptron network classifier includes three hidden layers, wherein the first hidden layer includes a linear transformation layer, a Dropout layer and a batch normalization layer, the second hidden layer includes a linear transformation layer and a ReLU activation function layer, and the third hidden layer includes a linear transformation layer.

[0058] Specifically, the specific process of inputting the global context vector into the multi-layer perceptron network classifier includes:

[0059] S61. Input the global context vector into the first hidden layer to obtain the first output, expressed as

[0060]

[0061] Where W1 represents the learnable weight matrix, b1 represents the learnable bias vector, c represents the global context vector, BatchNorm(·) represents the batch normalization layer, and Dropout(·) represents the Dropout layer. represents the first output;

[0062] S62. Input the first output into the second hidden layer to obtain the second output, expressed as

[0063]

[0064] in, represents the second output, ReLU(·) represents the ReLU activation function layer, W2 represents the learnable weight matrix, b2 represents the learnable bias vector,

[0065] S63. Input the second output into the third hidden layer to obtain the third output, expressed as

[0066]

[0067] Wherein, z3 represents the third output, W3 represents the learnable weight matrix, and b3 represents the learnable bias vector;

[0068] S64. Normalize the third output using the Sigmoid activation function to obtain a recognition result.

[0069] S7. Calculate the loss based on the recognition results and train the model parameters until the model parameters converge.

[0070] Specifically, the binary cross entropy loss function is used to effectively measure the error between the predicted value and the true label. The formula is:

[0071]

[0072] Where N represents the number of samples, y i represents the true label value of the i-th sample, Represents the model predicted label value of the i-th sample.

[0073] In the present invention, unless otherwise clearly stipulated and limited, the terms such as "installation", "setting", "connection", "fixation" and "rotation" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral one; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium; it can be the internal connection of two elements or the interaction relationship between two elements. Unless otherwise clearly defined, ordinary technicians in this field can understand the specific meanings of the above terms in the present invention according to the specific circumstances.

[0074] 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.

Claims

1. A method for identifying autism spectrum disorders based on time series, characterized in that: It includes building and training an autism recognition model, inputting the data to be recognized into the trained autism recognition model to obtain the recognition result; the autism recognition model includes a graph attention spatial network feature extractor, a bidirectional long-short term network temporal feature extractor and a multi-layer perceptron network classifier; The training process of the autism recognition model includes: S1. Acquire fMRI data and preprocess it to obtain preprocessed data; S2. Divide the whole brain into 200 brain regions according to the CC200 brain partition template, and extract the average time series data of each brain region based on the preprocessed data; S3. Calculate the Pearson correlation coefficient of each brain region based on the average time series data, and construct a graph structure based on the Pearson correlation coefficient; S4. Input the graph structure into the graph attention spatial network feature extractor to update the node features of each node; S5. Input the node features of the graph structure into the bidirectional long-short term network time feature extractor to obtain a global context vector; S6. Input the global context vector into the multi-layer perceptron network classifier to obtain the recognition result; S7. Calculate the loss based on the recognition results and train the model parameters until the model parameters converge.

2. The method for identifying autism spectrum disorders based on time series according to claim 1, characterized in that: The specific process of constructing the graph structure in step S3 includes: Each brain region is taken as a node, and the average time series data is used as the node feature; the Pearson correlation coefficient between each two nodes is calculated based on the time series data. If the Pearson correlation coefficient is greater than 0.5, an edge is connected between the two nodes, and the Pearson correlation coefficient is recorded as the edge value; if the Pearson correlation coefficient is not greater than 0.5, there is no connecting edge between the two nodes.

3. The method for identifying autism spectrum disorders based on time series according to claim 1, characterized in that: The graph attention spatial network feature extractor includes multiple attention graph convolution layers. The processing process of the lth attention graph convolution layer includes: S31. Take the output of node i in the -1th attention graph convolution layer As the input of the,th attention graph convolutional layer; S32. Input of the attention graph convolutional layer for node i The attention mechanism is used to perform neighbor node feature fusion processing to obtain the output of node i in the th attention graph convolution layer.

4. The method for identifying autism spectrum disorders based on time series according to claim 1, characterized in that: Step S5 specifically includes: S51. All node features are passed through a bidirectional LSTM network to obtain a comprehensive temporal representation at different time steps; S52. Calculate the attention score of the temporal comprehensive representation of each time step, expressed as e t =tanh(W a h t +b a ) Among them, e t represents the attention score at the tth time step, W a represents the learnable weight matrix, b a represents the learnable bias vector, h t represents the time-integrated representation of the t-th time step; S53. Normalize each attention score to get a scalar score, expressed as Among them, α t represents the scalar score of the tth time step, T represents the number of time steps; S54. A global context vector is obtained by weighted calculation based on the scalar score and the temporal comprehensive representation, which is expressed as Where c represents the global context vector.

5. The method for identifying autism spectrum disorders based on time series according to claim 1, characterized in that: The multilayer perceptron network classifier includes three hidden layers, wherein the first hidden layer includes a linear transformation layer, a Dropout layer, and a batch normalization layer, the second hidden layer includes a linear transformation layer and a ReLU activation function layer, and the third hidden layer includes a linear transformation layer.

6. The method for identifying autism spectrum disorders based on time series according to claim 5, characterized in that: The specific process of inputting the global context vector into the multilayer perceptron network classifier includes: S61. Input the global context vector into the first hidden layer to obtain the first output, expressed as Where W1 represents the learnable weight matrix, b1 represents the learnable bias vector, c represents the global context vector, BatchNorm(·) represents the batch normalization layer, and Dropout(·) represents the Dropout layer. represents the first output; S62. Input the first output into the second hidden layer to obtain the second output, expressed as in, represents the second output, ReLU(·) represents the ReLU activation function layer, W2 represents the learnable weight matrix, b2 represents the learnable bias vector, S63. Input the second output into the third hidden layer to obtain the third output, expressed as Wherein, z3 represents the third output, W3 represents the learnable weight matrix, and b3 represents the learnable bias vector; S64. Normalize the third output using the Sigmoid activation function to obtain a recognition result.

7. The method for identifying autism spectrum disorders based on time series according to claim 1, characterized in that: The loss is calculated using the binary cross entropy loss function, expressed as Where N represents the number of samples, y i represents the true label value of the i-th sample, Represents the model predicted label value of the i-th sample.

Images