ORIGINAL RESEARCH article
Front. Psychiatry
Sec. Neuroimaging
Integrating multi-atlas neuroimaging data for robust biomarker identification in neuropsychiatric disorders
Lijun Zhou 1,2
Yunfei Liu 1
Kaiyuan Wang 1
Xiaoyan Wang 1,2
Meirong He 1,2
Hongru Zhu 3,4
Junran Zhang 1
1. Sichuan University College of Electrical Engineering, Chengdu, China
2. Northwest Minzu University, Lanzhou, China
3. Sichuan University West China Hospital Mental Health Center, Chengdu, China
4. Med-X Center for Informatics, Sichuan University, Chengdu, China, Chengdu, China
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Abstract
Accurate diagnosis of neuropsychiatric disorders such as autism spectrum disorder (ASD) and post-traumatic stress disorder (PTSD) remains challenging due to complex disruptions in brain functional connectivity, particularly within frontal-temporal networks. Existing graph neural network (GNN)-based methods often rely on single-atlas representations, limiting their ability to capture cross-atlas complementary information and subtle network abnormalities. To address these limitations, we propose MSAT-LAFFNet, an improved GNN-based classification framework that integrates multi-atlas features with a structure-aware graph Transformer (SAT) and a lightweight attentional feature fusion network (LAFFNet). The SAT module enhances structural encoding by incorporating graph topology into attention computations, while LAFFNet adaptively fuses cross-atlas features to reduce redundancy and strengthen disease-discriminative representations. The framework was validated on the publicly available ABIDE-I dataset and a private PTSD dataset (n=138) from West China Hospital of Sichuan University. Our model achieved an AUC of 82.9% and an accuracy of 81.59% on ABIDE-I, and an AUC of 89.96% with 89.45% accuracy on the PTSD dataset, outperforming competing models. Additionally, The proposed method enhances clinical interpretability by identifying overlapping abnormal brain regions, which in ASD include the right middle frontal gyrus (MFG.R) and the left inferior temporal gyrus (ITG.L), and in PTSD include the left middle temporal gyrus (MTG.L), the right medial superior frontal gyrus (SFGmed.R) and the left amygdala (AMYG.L), and by elucidating key mechanisms such as weakened prefrontal-temporal connectivity in ASD and default mode network (DMN)-amygdala decoupling in PTSD. These findings demonstrate MSAT-LAFFNet's potential as an effective tool for auxiliary diagnosis of neuropsychiatric disorders characterised by disrupted functional networks.
Summary
Keywords
Autism Spectrum Disorder, deep learning, functional connectivity, Graph neural network, multi-atlas, Post-traumatic stress disorder
Received
12 October 2025
Accepted
20 February 2026
Copyright
© 2026 Zhou, Liu, Wang, Wang, He, Zhu and Zhang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Hongru Zhu; Junran Zhang
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.