Search

KR-20260064762-A - APPARATUS AND METHOD FOR DIAGNOSING AUTISM SPECTRUM DISORDER(ASD) USING MULTI-HEAD ATTENTION-BASED DYNAMIC FUNCTIONAL CONNECTIVITY

KR20260064762AKR 20260064762 AKR20260064762 AKR 20260064762AKR-20260064762-A

Abstract

The present invention relates to a graph neural network-based autism spectrum disorder (ASD) diagnostic device comprising: a preprocessing unit that acquires brain imaging data and designates a Region of Interest (ROI) to generate preprocessed data suitable for neural network input; a spatial feature extraction unit that extracts spatial features from the preprocessed data; a temporal feature extraction unit that analyzes changes in brain activity over time and extracts attention-based temporal features; a spatiotemporal fusion feature unit that combines spatial and temporal features to analyze spatiotemporal correlations; a graph generation unit that converts connectivity between each region of interest into a graph structure and implements it as nodes and edges; and a graph classification unit that analyzes spatiotemporal patterns of connectivity through the graph structure and classifies whether or not there is autism spectrum disorder (ASD).

Inventors

  • 조성배
  • 문형준

Assignees

  • 연세대학교 산학협력단

Dates

Publication Date
20260508
Application Date
20241028

Claims (10)

  1. A preprocessing unit that acquires brain imaging data and designates a region of interest (ROI) of the brain to generate preprocessing data for neural network input; A spatial feature extraction unit that extracts spatial features for the region of interest of the brain from the above preprocessed data; A time feature extraction unit that derives weighted changes in brain activity over time for the region of interest of the brain for the above-mentioned preprocessed data and extracts attention-based temporal features; A spatiotemporal fusion feature unit that analyzes the spatiotemporal correlation of brain activity by combining the above spatial feature and the above attention-based temporal feature; A graph generation unit that represents the connectivity between each region of interest in the brain as a graph structure, wherein the graph structure implements each region of interest as nodes and the connectivity as edges; and A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, comprising a graph classification unit that classifies whether there is autism spectrum disorder (ASD) by analyzing the spatiotemporal pattern of connectivity through the graph structure.
  2. In paragraph 1, the pretreatment unit A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by acquiring fMRI data using the brain imaging data and acquiring the region of interest of the brain as multiple window unit images through a sliding window to generate the preprocessed data.
  3. In claim 1, the spatial feature extraction unit A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by performing CNN operations on the above-mentioned preprocessed data to generate a feature map representing the degree of activation or activity pattern of the above-mentioned region of interest as the above-mentioned spatial feature.
  4. In claim 1, the time feature extraction unit A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by applying a bidirectional LSTM (Long Short-Term Memory) network to the above-mentioned preprocessed data to generate temporal features of previous or subsequent activities that occurred in the region of interest.
  5. In paragraph 4, the time feature extraction unit A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by applying a multi-head attention mechanism to the above temporal features to generate attention representing the temporal interaction of the above activities.
  6. In paragraph 5, the time feature extraction unit A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by integrating the above temporal feature and the above attention and performing Global Average Pooling to summarize the weighted change of the above brain activity into the above attention-based temporal feature.
  7. In claim 1, the space-time fusion feature part A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by generating a correlation matrix representing the degree of spatiotemporal interaction of brain activity to represent spatiotemporal correlation by combining the spatial feature and the attention-based temporal feature.
  8. In paragraph 1, the graph generating unit A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by determining the weight of the edge by quantifying the connectivity between each region of interest of the brain through a correlation matrix derived by the spatiotemporal fusion feature unit.
  9. In paragraph 1, the graph classification unit A diagnostic device for autism spectrum disorder (ASD) using a graph neural network, characterized by generating graph embeddings for the above graph structure and classifying whether the autism spectrum disorder (ASD) is present through a fully connected layer and a softmax layer.
  10. In a method for diagnosing autism spectrum disorder (ASD) using a graph neural network, performed in an autism spectrum disorder (ASD) diagnostic device using a graph neural network, A preprocessing step of acquiring brain imaging data and designating the brain's Region of Interest (ROI) to generate preprocessing data for neural network input; A spatial feature extraction step for extracting spatial features for the region of interest of the brain from the above preprocessed data; A temporal feature extraction step for extracting attention-based temporal features by deriving weighted changes in brain activity over time for the region of interest of the brain regarding the above preprocessed data; A spatiotemporal fusion feature step that analyzes the spatiotemporal correlation of brain activity by combining the above spatial feature and the above attention-based temporal feature; A graph generation step that represents the connectivity between each region of interest in the brain as a graph structure, wherein the graph structure implements each region of interest as nodes and the connectivity as edges; and A method for diagnosing autism spectrum disorder (ASD) using a graph neural network, comprising a graph classification step that classifies whether there is autism spectrum disorder (ASD) by analyzing the spatiotemporal pattern of connectivity through the graph structure.

Description

Apparatus and Method for Diagnosing Autism Spectrum Disorder Using a Graph Neural Network The present invention relates to a diagnostic technology for autism spectrum disorder (ASD), and more specifically, to an apparatus and method for diagnosing autism spectrum disorder using a graph neural network that can improve the accuracy of autism spectrum disorder diagnosis by combining a multi-head attention technique and a graph neural network (GNN). Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by diverse developmental progressions and neuroanatomical features among patients, making diagnosis a complex and challenging task. While deep learning models based on functional magnetic resonance imaging (fMRI) have recently shown promising results, existing studies have primarily focused on overall global activation patterns, failing to adequately capture local characteristics and limiting their ability to accurately assess signs of the disease. While methods utilizing functional connectivity (FC) are useful for modeling functional relationships in the brain, many existing methods focus solely on structural features or fail to consider the heterogeneous external characteristics of the data, resulting in reduced efficiency in ASD diagnosis and the oversight of transient and localized characteristics. Furthermore, since some symptoms may manifest intensively only at specific times and in localized areas, it is crucial to effectively capture these characteristics. Korean Published Patent No. 10-2023-0024667 (February 21, 2023) provides a 4D fMRI autism prediction and early diagnosis method and device that enables accurate prediction and early diagnosis of autism by providing a model that learns patient-specific spatial variability factors while minimizing information loss of time-series features and extracting autism characteristics through weighting of time-series connectivity. The autism prediction and diagnosis device can accurately predict and early diagnose autism by outputting autism characteristics through an autism prediction and diagnosis model that includes a spatial structure model of brain regions and a dynamic time-series connectivity model of brain regions using time-series brain imaging data. Figure 1 is a flowchart illustrating the operation process of an autism spectrum disorder (ASD) diagnostic device using a graph neural network according to the present invention. FIG. 2 is a diagram illustrating the configuration of an autism spectrum disorder (ASD) diagnostic device according to the present invention. Figure 3 is a flowchart illustrating the operation of the autism spectrum disorder (ASD) diagnostic device of Figure 2. Figure 4 is a box plot visually comparing the diagnostic accuracy of the autism spectrum disorder (ASD) diagnostic method according to the present invention with that of several existing models. Figure 5 is a graph showing the Receiver Operating Characteristic (ROC) curve evaluating the performance of the autism spectrum disorder (ASD) diagnostic device according to the present invention. Figure 6 is a brain scan image visually showing the results of region of interest (ROI) activation in the brain analyzed by the autism spectrum disorder (ASD) diagnostic device according to the present invention. Figure 7 is a visual representation of the analysis results of an autism spectrum disorder (ASD) diagnostic device according to the present invention, and is a diagram visually representing a correlation matrix and a graph of connections between brain regions that compare functional connectivity appearing in the brains of a patient with autism spectrum disorder (ASD) and a normal control group. Figure 8 is a diagram showing the analysis results of an autism spectrum disorder (ASD) diagnostic device according to the present invention, which visually compares the functional brain connectivity of an autism spectrum disorder (ASD) patient and a normal control group. The description of the present invention is merely an example for structural or functional explanation, and therefore the scope of the present invention should not be interpreted as being limited by the examples described in the text. That is, since the examples are subject to various modifications and may take various forms, the scope of the present invention should be understood to include equivalents capable of realizing the technical concept. Furthermore, the objectives or effects presented in the present invention do not imply that a specific example must include all of them or only such effects; therefore, the scope of the present invention should not be understood as being limited by them. Meanwhile, the meaning of the terms described in this application should be understood as follows. Terms such as "first," "second," etc., are intended to distinguish one component from another, and the scope of rights shall not be limited by these terms. For example, the first component may be named the second