Cross-attention-guided subject-adaptive graph learning for multimodal autism classification: integrating structural and functional MRI data

Multi-Scale Structural MRI Features Reveal Task-Based Functional Connectivity and Its Alterations in Psychiatric Disorders: A Collaborative Graph Attention Network Approach

Enhanced multimodal MRI classification of schizophrenia through cross-attention graph neural networks.

We propose TB-GCAN, a tri-branch cross-attention graph neural network for schizophrenia classification using multimodal MRI, including sMRI, fMRI, and DTI. Built on a multi-site dataset of 1191 samples from seven scanning sites, the model exploits atlas-defined one-to-one anatomical correspondence across modalities to enable node-level cross-attention during intermediate representation learning. In 7-site leave-one-site-out evaluation, TB-GCAN achieved 84.63% accuracy and outperformed GAT, GCN, CNN, SVM, and MMGNN in the tri-modal setting. Attention-based region ranking highlighted biologically plausible schizophrenia-related regions, and downstream analyses linked the learned imaging representations to PANSS dimensions and transcriptional programs. Unlike generic multimodal GNNs that learn cross-modal relations from data, TB-GCAN directly leverages atlas-aligned regional correspondence to perform anatomically constrained node-level interaction. These findings indicate that anatomically grounded node-level multimodal fusion can improve classification performance while preserving neurobiological interpretability, thereby providing a principled framework for multimodal schizophrenia classification and biomarker discovery.

Multi-head attention-driven multimodal feature integration network for autism spectrum disorder detection

The growing prevalence of Autism Spectrum Disorder (ASD) highlights the need for accurate and reliable intelligent screening systems for early behavioral assessment. However, ASD-related behaviors vary significantly across individuals, making diagnosis based on a single modality or subjective evaluation unreliable. Moreover, existing computational approaches often struggle to effectively model complex spatial and temporal dependencies in behavioral video data, leading to limited feature representation. To address these challenges, this study proposes a Multi-head Attention-driven Multimodal Feature Integration Network (MAtMFIN) that jointly analyzes Eye-Tracking (ET) scanpath data and behavioral video data to improve the robustness of ASD screening. The framework employs a Hierarchical Attention Refinement block (HARb) and a Spatial Enhancement Module (SEM) for effective feature refinement, along with a cross-modal cross-attention mechanism to capture complementary relationships between modalities. Experimental evaluation on the proposed multimodal ASD dataset demonstrates that the proposed MAtMFIN framework consistently outperforms state-of-the-art transfer learning and Deep Learning (DL) models, including a hybrid Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) model, 2D-CNN with attention, LSTM-attention, Inflated 3D CNN, and Spatio-Temporal Graph Neural Networks (STGNN). The proposed method achieves a training accuracy of 94.6%, a testing accuracy of 93.2%, and an F1 score of 93.5%. These results indicate that effective cross-modal feature integration and attention-based modelling of spatio-temporal dependencies significantly enhance the ASD behavioral analysis, highlighting the potential of the proposed framework for intelligent healthcare applications.

Detection of ASD Using DySTEP-GAT Architecture With EEG-Based Attention-Fused Brain-Net Graph Representation

Findings indicate that the learned brain network representations capture distinct and quantifiable abnormalities in ASD-related neural dynamics, establishing the proposed framework as a robust, interpretable, and measurement-aligned tool for objective ASD assessment.

Multimodal neuroimaging fusion with hierarchical structure-function coupling for autism spectrum disorder diagnosis.

Multimodal neuroimaging fusion has gained considerable attention in diagnosing autism spectrum disorder (ASD) due to its ability to integrate complementary information across different modalities. However, most existing approaches rely on simple feature concatenation or late fusion strategies without achieving effective cross-modal coupling, failing to capture the intrinsic relationships between functional and structural information. To address these limitations, this study develops the deep multimodal dual-level coupling (DMDC) framework integrating brain functional and structural information to improve model discrimination capability and interpretability. Specifically, the pyramid-inverted graph-attention network is first proposed to dynamically update graph topological structures with the Top-k node filtering algorithm for function-structure network coupling. Second, the skeleton-based white matter projection method maps functional magnetic resonance imaging (fMRI) signals onto diffusion tensor imaging (DTI) derived white matter skeletons, followed by the multi-layered hierarchical convolutional network for representation extraction. Finally, the weighted feature integration mechanism combines both components, and the neural network with cross-entropy loss optimization is employed for ASD identification. Extensive experiments show DMDC outperforms state-of-the-art methods. Key discriminative regions identified include the hippocampus, anterior cingulate cortex, amygdala, and frontal gyri. Both hyperconnectivity and hypoconnectivity were observed in ASD, especially in prefrontal cortex, amygdala, and hippocampus, which were critical for social and emotional processing. The proposed framework demonstrates robust diagnostic performance and provides reliable biomarkers for neurological assessment. The identified regions and connectivity patterns offer insights into ASD neural mechanisms and potential diagnostic biomarkers.

Hierarchical neighbor integration graph attention network for autism spectrum disorder diagnosis

Introduction Early diagnosis of autism spectrum disorder (ASD) based on resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for effective intervention and rehabilitation. Using rs-fMRI, functional brain networks (FBNs) are constructed to represent interactions among brain regions of interest (ROIs), and existing graph neural network–based methods, particularly Graph Attention Networks (GATs), have shown promise for FBN-based ASD diagnosis. However, most current approaches primarily aggregate ROIs through low-order pairwise interactions, while higher-order neighbors are incorporated only implicitly through increased network depth. This strategy often leads to over-smoothing of node representations and limits the capture of informative higher-order brain interactions. Methods To address these challenges, we propose the Hierarchical Neighbor Integration Graph Attention Network (HiNIGAT), a general graph learning framework that explicitly models multi-order interactions in FBNs. Specifically, HiNIGAT introduces a multi-order attention mechanism that constrains each attention head to specialize in a distinct neighborhood order, enabling the model to capture brain interactions from local connectivity to global network structure. Furthermore, a bidirectional gated fusion strategy is proposed to adaptively integrate complementary information across multi-order features, facilitating effective local–global representation collaboration. Results Experiments on the ABIDE-I dataset demonstrate the effectiveness of HiNIGAT. Discussion The results highlight the importance of explicit multi-order integration for ASD diagnosis.