Autism spectrum disorder (ASD) is a serious neurodevelopmental disorder for children and adolescent. Accurate diagnosis of ASD plays a key role in improving the life quality of individuals with ASD and reducing the burden of healthcare system. Currently, Functional Connectivity (FC) analysis based on functional Magnetic Resonance Imaging (fMRI) has become a popular approach in diagnosis of ASD, but majority of previous studies were based on the assumption that FC is stationary throughout the entire scan session, ignoring the fluctuations over the course of the scan. Previously sliding window based Dynamic FC (DFC) method was proposed to estimate the dynamic changes, but it has a limitation that all observations within the window are weighted equally. To address the issue, we have proposed Dynamic Weighted FC (DWFC) method in this study to extract features from resting-state fMRI (rs-fMRI) and applied it to distinguishing ASD patients from Normal Controls (NC). Experiments were carried on subjects from the Autism Brain Imaging Data Exchange (ABIDE) database. The classification accuracy is 0.8525 and 0.8061 on two independent datasets. Results showed that the proposed method significantly outperformed conventional FC and DFC approaches, as well as other state-of-the-art ASD classification methods, which suggests this method as a promising computer-acid diagnosis tool for ASD.
Hongliang Zou and Jian Yang