MCDGLN: Masked Connection-based Dynamic Graph Learning Network for Autism Spectrum Disorder

TL;DR

The paper tackles ASD classification from rs-fMRI by modeling both dynamic and static brain connectivity. It introduces MCDGLN, which combines a weighted edge aggregation of dynamic functional connectivity with masked edge drop, a hierarchical graph convolutional network, and an attention-based encoder to produce robust, task-relevant representations for classification. Key contributions include the WEA module for dynamic-static fusion, the MED mechanism for noise pruning, and the ACE-GCN–SA pipeline that yields state-of-the-art accuracy (notably 73.3% on ABIDE-I CC200) and interpretable connectivity patterns. The approach advances ASD biomarker discovery by more accurately capturing dynamic brain interactions and pruning irrelevant connections, with potential impact on clinical diagnostics and neuroscience research.

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by complex physiological processes. Previous research has predominantly focused on static cerebral interactions, often neglecting the brain's dynamic nature and the challenges posed by network noise. To address these gaps, we introduce the Masked Connection-based Dynamic Graph Learning Network (MCDGLN). Our approach first segments BOLD signals using sliding temporal windows to capture dynamic brain characteristics. We then employ a specialized weighted edge aggregation (WEA) module, which uses the cross convolution with channel-wise element-wise convolutional kernel, to integrate dynamic functional connectivity and to isolating task-relevant connections. This is followed by topological feature extraction via a hierarchical graph convolutional network (HGCN), with key attributes highlighted by a self-attention module. Crucially, we refine static functional connections using a customized task-specific mask, reducing noise and pruning irrelevant links. The attention-based connection encoder (ACE) then enhances critical connections and compresses static features. The combined features are subsequently used for classification. Applied to the Autism Brain Imaging Data Exchange I (ABIDE I) dataset, our framework achieves a 73.3\% classification accuracy between ASD and Typical Control (TC) groups among 1,035 subjects. The pivotal roles of WEA and ACE in refining connectivity and enhancing classification accuracy underscore their importance in capturing ASD-specific features, offering new insights into the disorder.

Figures (7)

• Figure 1: The schematic diagram of our proposed framework. The input is the BOLD signal, which is separated by sliding windows to form diversity dFCs for each subject and the output is the prediction of the model. (a) is the section for the process of extracting and compressing the static features and (b) is that of distilling and squeezing the dynamic features.
• Figure 2: The difference between normal convolution and cross convolution. The kernel in the cross convolution is channel-wise and element-wise.
• Figure 3: The results of ablation study on CC200 atlas
• Figure 4: The results of ablation study on AAL atlas
• Figure 5: The influence of the WEA's layers on two atlases