Augmentation-based Unsupervised Cross-Domain Functional MRI Adaptation for Major Depressive Disorder Identification

TL;DR

This work addresses cross-site variability in rs-fMRI-based MDD identification by introducing AUFA, an augmentation-based unsupervised cross-domain adaptation framework. AUFA combines Transformer-based graph representation learning of functional connectivity networks with a maximum mean discrepancy-based domain alignment and a target-domain augmentation strategy to mitigate overfitting and improve generalization to unseen sites. The method yields superior MDD/NC classification across four cross-site tasks, outperforming both traditional and deep-domain-adaptation baselines, and additionally highlights disease-relevant brain connections for interpretability. The approach advances multi-site neuroimaging diagnostics by reducing site heterogeneity while providing actionable biomarkers tied to discriminative connectivity patterns.

Abstract

Major depressive disorder (MDD) is a common mental disorder that typically affects a person's mood, cognition, behavior, and physical health. Resting-state functional magnetic resonance imaging (rs-fMRI) data are widely used for computer-aided diagnosis of MDD. While multi-site fMRI data can provide more data for training reliable diagnostic models, significant cross-site data heterogeneity would result in poor model generalizability. Many domain adaptation methods are designed to reduce the distributional differences between sites to some extent, but usually ignore overfitting problem of the model on the source domain. Intuitively, target data augmentation can alleviate the overfitting problem by forcing the model to learn more generalized features and reduce the dependence on source domain data. In this work, we propose a new augmentation-based unsupervised cross-domain fMRI adaptation (AUFA) framework for automatic diagnosis of MDD. The AUFA consists of 1) a graph representation learning module for extracting rs-fMRI features with spatial attention, 2) a domain adaptation module for feature alignment between source and target data, 3) an augmentation-based self-optimization module for alleviating model overfitting on the source domain, and 4) a classification module. Experimental results on 1,089 subjects suggest that AUFA outperforms several state-of-the-art methods in MDD identification. Our approach not only reduces data heterogeneity between different sites, but also localizes disease-related functional connectivity abnormalities and provides interpretability for the model.

Figures (5)

• Figure 1: Overview of our augmentation-based unsupervised cross-domain fMRI adaptation (AUFA) framework, including (1) fMRI feature extraction for brain functional connectivity networks using a Transformer module, (2) domain adaptation module for feature alignment between source and target domains, (3) augmentation-based self-optimization module, and (4) classification module. The data from the target domain are fed into the model that has been jointly optimized by $L_{C}$, $L_{M}$ and $L_{A}$ for direct prediction.
• Figure 2: Graph representation learning module in AUFA that learns new features with spatial attention through multiple Transformer encoder layers.
• Figure 3: Visualization of feature distributions at different sites before and after domain adaptation by t-SNE van2008visualizing, where yellow "+" denotes the source domain and green dots denote the target domains.
• Figure 4: Accuracy (ACC) achieved by AUFA in MDD and NC classification with different values of $\lambda_{1}$ and $\lambda_{2}$.
• Figure 5: Visualization of the top ten most discriminative functional connections detected by AUFA in the MDD vs. NC classification task using the BrainNet Viewer xia2013brainnet.