3D-Resnet Fused Attention for Autism Spectrum Disorder Classification

Abstract

Autism spectrum disorder (ASD) is a neuro development condition, early diagnosis of ASD traits is indispensable for applying effective treatment. Analysis of ASD based on deep learning (DL) has become an active research topic in the field of medical image analysis. However, the classification performance of the DL models is severely affected by the class imbalance of medical images. Recently, the combination of attention mechanism and Convolutional Neural Network (CNN) has significantly improved the performance of numerous classification and recognition tasks. Therefore, the attention mechanism has a promising prospect in improving the performance of CNN-based ASD diagnosis. This paper proposes a 3D-ResNet model with an attention subnet for ASD diagnosis. The model is constructed by the residual attention module, which is designed to mask redundant regions that are irrelevant to ASD diagnostic classification during the feature extraction process. The fMRI from Autism Imaging Data Exchange (ABIDE) was applied as the datasets and the experimental results show that the performance of the proposed approach is significantly better than that of conventional DL models.