Autism spectrum disorder identification system and method based on graph neural network

Abstract

The invention discloses an infantile autism spectrum disorder identification system based on a graph neural network, and the system specifically comprises a data processing module which is used for extracting time sequence information from fMRI data and constructing a brain function connection graph; the model training module is configured with an unsupervised graph auto-encoder GAE and comprises a three-layer graph convolutional network GCN as an encoder and an inner product decoder; the biomarker recognition module is used for executing a replacement test and screening brain regions with significant differences; and the performance evaluation module is used for evaluating the diagnostic performance of the system and the effectiveness of biomarker discovery on the ABIDE data set. The invention provides an ASD identification solution which does not need manual annotation, is high in interpretability and excellent in performance, and has important significance for promoting the development of ASD early diagnosis and personalized treatment.

Description

Technical Field

[0001] The present invention relates to the field of medical diagnosis and processing technology, and in particular to an autism spectrum disorder identification system and method based on a graph neural network. Background Art

[0002] The diagnosis of autism spectrum disorder (ASD) has long relied on behavioral assessments and questionnaires. These methods are highly dependent on clinician experience and patient expressiveness, resulting in significant variability in diagnostic consistency and accuracy. Furthermore, while functional magnetic resonance imaging (fMRI) data provides rich information for understanding brain structure and function, traditional linear models face challenges in processing this data, struggling to capture the complex nonlinear interactions between brain regions. While existing deep learning methods such as convolutional neural networks (CNNs) have achieved significant success in image recognition, they perform poorly when processing non-Euclidean data such as brain connectivity maps, limiting their application in imaging-based ASD identification. Furthermore, these methods often require extensive amounts of labeled data for training. In the context of ASD diagnosis, obtaining high-quality, clearly labeled fMRI datasets is a significant challenge. Consequently, existing technical approaches have significant shortcomings in reducing reliance on scarce labeled data and improving diagnostic specificity and accuracy. In particular, there is a lack of effective solutions for identifying significantly differentially expressed brain regions and their correlation with ASD symptom severity. Summary of the Invention

[0003] The present invention aims to solve the above-mentioned problems and provide an autism spectrum disorder identification system and method based on graph neural network.

[0004] To solve the above technical problems, the present invention provides a technical solution: an autism spectrum disorder identification system based on a graph neural network, the system specifically comprising:

[0005] Data processing module, used to extract time series information from fMRI data and construct brain functional connectivity maps;

[0006] The model training module is configured with an unsupervised graph autoencoder (GAE), including a three-layer graph convolutional network (GCN) as an encoder and an inner product decoder;

[0007] A biomarker identification module, used to perform permutation tests and screen brain regions with significant differences;

[0008] The performance evaluation module is used to evaluate the diagnostic performance of the system and the effectiveness of biomarker discovery on the ABIDE dataset.

[0009] As an improvement, the system also includes a visualization module that can display the identified differential brain regions to improve the interpretability of the model.

[0010] As an improvement, an identification method for an autism spectrum disorder identification system based on a graph neural network is provided, the method specifically comprising the following steps:

[0011] S1. Extract time series information from functional magnetic resonance imaging (fMRI) data and construct a brain functional connectivity map, where nodes represent brain regions and edges represent Pearson correlation coefficients.

[0012] S2. Use the AAL template to divide the brain into 90 regions and generate a dataset of images of the ASD group and the control group.

[0013] S3. Apply an unsupervised graph autoencoder (GAE). The encoder contains a three-layer graph convolutional network (GCN) to extract node embeddings. The dimension is set to 128, and the decoder uses the inner product method to reconstruct the adjacency matrix.

[0014] S4. During the training process, reconstruction loss plus contrast loss is used as the total loss function to enhance the differences between groups;

[0015] S5. Perform a permutation test on the two sets of node embeddings, with 1000 iterations and a p-value less than 0.05 after FDR correction, to screen out brain regions with significant differences.

[0016] As an improvement, the method further includes biomarker validation of the identified brain regions, particularly Vermis_3 / Vermis_4_5 and fusiform gyrus of the cerebellum, and evaluating the correlation between these brain regions and the severity of ASD symptoms.

[0017] As an improvement, the AUC value of ASD classification reached 0.89 on the ABIDE dataset, and the importance of known ASD-related brain regions such as the temporal lobe and occipital lobe was verified. At the same time, a significant correlation between the Vermis area of the cerebellum and the severity of ASD symptoms was reported for the first time.

[0018] The advantages of the present invention compared with the prior art are:

[0019] The graph neural network-based autism spectrum disorder identification system and method proposed in this paper achieves efficient and accurate identification of autism spectrum disorder (ASD) by innovatively applying an unsupervised graph autoencoder (GAE) to process functional magnetic resonance imaging (fMRI) data. This system not only overcomes the limitations of traditional linear models in capturing complex nonlinear interactions between brain regions, but also reduces the reliance on scarce annotated data, solving a major challenge in ASD diagnosis. Specifically, by constructing a brain functional connectivity map and extracting features using a three-layer graph convolutional network (GCN), the present invention can effectively identify brain regions with significant differences, such as Vermis_3 / Vermis_4_5 in the cerebellum and the fusiform gyrus, and for the first time reports a significant correlation between these regions and ASD symptom severity. In addition, experimental results on the ABIDE dataset show that the present method improves the AUC value by 24% to 0.89 compared to traditional CNN, confirming the importance of known ASD-related brain regions such as the temporal lobe and occipital lobe, further improving diagnostic specificity and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a structural diagram of the autism spectrum disorder identification system based on graph neural network of the present invention;

[0021] Figure 2 This is a flow chart of an identification method of an autism spectrum disorder identification system based on a graph neural network of the present invention; DETAILED DESCRIPTION

[0022] In the description of the present invention, it should be understood that the terms "up", "down", "front", "back", "left", "right", "inside", "outside", "center", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction structure and operation, and therefore cannot be understood as a limitation on the present invention.

[0023] In the description of the present invention, it should also be noted that, unless otherwise expressly specified or limited, the terms "provided with," "installed," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections, electrical connections, direct connections, indirect connections via an intermediate medium, and internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0024] The present invention will be described in further detail below with reference to the accompanying drawings.

[0025] Combined with attachment Figure 1 —Attachment Figure 2 , an autism spectrum disorder identification system based on graph neural network, the system specifically includes:

[0026] Data processing module, used to extract time series information from fMRI data and construct brain functional connectivity maps;

[0027] The model training module is configured with an unsupervised graph autoencoder (GAE), including a three-layer graph convolutional network (GCN) as an encoder and an inner product decoder;

[0028] A biomarker identification module, used to perform permutation tests and screen brain regions with significant differences;

[0029] The performance evaluation module is used to evaluate the diagnostic performance of the system and the effectiveness of biomarker discovery on the ABIDE dataset.

[0030] The visualization module can display the identified differential brain regions and improve the interpretability of the model.

[0031] A method for identifying an autism spectrum disorder (ASD) system based on a graph neural network, the method specifically comprising the following steps:

[0032] S1. Extract time series information from functional magnetic resonance imaging (fMRI) data and construct a brain functional connectivity map, where nodes represent brain regions and edges represent Pearson correlation coefficients.

[0033] S2. Use the AAL template to divide the brain into 90 regions and generate a dataset of images of the ASD group and the control group.

[0034] S3. Apply an unsupervised graph autoencoder (GAE). The encoder contains a three-layer graph convolutional network (GCN) to extract node embeddings. The dimension is set to 128, and the decoder uses the inner product method to reconstruct the adjacency matrix.

[0035] S4. During the training process, reconstruction loss plus contrast loss is used as the total loss function to enhance the differences between groups;

[0036] S5. Perform a permutation test on the two sets of node embeddings, with 1000 iterations and a p-value less than 0.05 after FDR correction, to screen out brain regions with significant differences.

[0037] The working principle of the present invention is: the system of the present invention is composed of multiple modules to ensure the automation and efficiency of the entire process from data processing to diagnostic performance evaluation. First, the data processing module is responsible for extracting time series information from functional magnetic resonance imaging (fMRI) data and constructing a brain functional connection map, in which nodes represent brain regions and edges represent Pearson correlation coefficients. The brain is divided into 90 regions using the AAL template to generate graph data sets for the ASD group and the control group. Next, the model training module is configured with an unsupervised graph autoencoder (GAE), which contains a three-layer graph convolutional network (GCN) as an encoder for extracting node embeddings (the dimension is set to 128), and the decoder reconstructs the adjacency matrix by inner product. During the training process, the reconstruction loss plus the contrast loss is used as the total loss function to enhance the differences between groups. Subsequently, the biomarker identification module performed a permutation test on the two sets of node embeddings (1000 iterations, p-value less than 0.05 and FDR correction), screening out brain regions with significant differences, especially areas such as the cerebellum Vermis_3 / Vermis_4_5 and the fusiform gyrus, and further verified the correlation between these brain regions and the severity of ASD symptoms. In addition, this system is also equipped with a visualization module that can display the identified differential brain regions, greatly improving the interpretability of the model. Finally, the performance evaluation module evaluated the diagnostic performance of the system and the effectiveness of biomarker discovery on the ABIDE dataset, achieving an AUC value of 0.89 for ASD classification, and verifying the importance of known ASD-related brain regions such as the temporal lobe and occipital lobe. At the same time, it reported for the first time a significant correlation between the cerebellum Vermis area and the severity of ASD symptoms. This implementation method not only improves the accuracy and specificity of ASD diagnosis, but also provides a powerful tool for exploring new biomarkers.

[0038] The present invention and its embodiments are described above. This 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 a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs a structure similar to the technical solution without creatively designing an embodiment, all of which shall fall within the scope of protection of the present invention.

Claims

1. Autism spectrum disorder identification system based on graph neural network, characterized by: The system specifically includes: Data processing module, used to extract time series information from fMRI data and construct brain functional connectivity maps; The model training module is configured with an unsupervised graph autoencoder (GAE), including a three-layer graph convolutional network (GCN) as an encoder and an inner product decoder; A biomarker identification module, used to perform permutation tests and screen brain regions with significant differences; The performance evaluation module is used to evaluate the diagnostic performance of the system and the effectiveness of biomarker discovery on the ABIDE dataset.

2. The autism spectrum disorder identification system based on graph neural network according to claim 1 is characterized in that .The system also includes a visualization module that can display the identified differential brain regions to improve the interpretability of the model.

3. A method for identifying autism spectrum disorder based on a graph neural network according to claim 1, characterized in that: The method specifically comprises the following steps: S1. Extract time series information from functional magnetic resonance imaging (fMRI) data and construct a brain functional connectivity map, where nodes represent brain regions and edges represent Pearson correlation coefficients. S2. Use the AAL template to divide the brain into 90 regions and generate a dataset of images of the ASD group and the control group. S3. Apply an unsupervised graph autoencoder (GAE). The encoder contains a three-layer graph convolutional network (GCN) to extract node embeddings. The dimension is set to 128, and the decoder uses the inner product method to reconstruct the adjacency matrix. S4. During the training process, reconstruction loss plus contrast loss is used as the total loss function to enhance the differences between groups; S5. Perform a permutation test on the two sets of node embeddings, with 1000 iterations and a p-value less than 0.05 after FDR correction, to screen out brain regions with significant differences.

4. The identification method of the autism spectrum disorder identification system based on graph neural network according to claim 3, characterized in that: The method further includes performing biomarker validation on the identified brain regions, particularly Vermis_3 / Vermis_4_5 and fusiform gyrus of the cerebellum, and evaluating the correlation between these brain regions and the severity of ASD symptoms.

5. The identification method of the autism spectrum disorder identification system based on graph neural network according to claim 3, characterized in that: The AUC value of ASD classification reached 0.89 on the ABIDE dataset, and the importance of known ASD-related brain regions such as the temporal lobe and occipital lobe was verified. At the same time, a significant correlation between the Vermis area of the cerebellum and the severity of ASD symptoms was reported for the first time.

Images