ADHD fMRI classification through CBAM-Autoencoder framework

Abstract

This work presents a novel method for leveraging resting-state functional magnetic resonance imaging (fMRI) data to accurately detect Attention Deficit Hyperactivity Disorder (ADHD). The proposed method integrates the Convolutional Block Attention Module (CBAM) with a lightweight Autoencoder network to effectively extract and highlight salient features within fMRI scans. By leveraging attention mechanisms, the model focuses on important local details while filtering out irrelevant information, thereby enhancing diagnostic precision. Extensive experimentation on the ADHD-200 dataset showcases the efficacy of the proposed approach, demonstrating its ability to improve classification performance significantly. Specifically, the method achieved an average accuracy of 91.7% across the NYU, 93.8% across the KKI, 86.4% across the NI, 89.1% across the PU, and 83.5% across the PU_1 datasets. This research underscores the potential of attention-based deep learning techniques in advancing ADHD diagnosis using neuroimaging data.