Sparse Low-rank Constrained Adaptive Structure Learning using Multi-template for Autism Spectrum Disorder Diagnosis

Abstract

Autism spectrum disorder (ASD) is a developmental disability that causes severe social, communication and behavioral challenges. Up to now, many imaging-based approaches for ASD diagnosis have been proposed. However most of them limited to single template. In this paper, we propose a novel sparse low-rank constrained multi-templates data based method for ASD diagnosis, which performs feature selection and adaptive local structure learning simultaneously. Specifically, we encode modularity prior while constructing functional connectivity (FC) brain networks from different templates for each subject. After extracting features from FC networks, feature selection is applied. Meanwhile, the local structure is learnt via an adaptive process. Extensive experiments are conducted to demonstrate the effectiveness of our proposed method on the Autism Brain Imaging Data Exchange (ABIDE) database. Experimental results verify our proposed method can enhance the diagnosis performances and outperform the commonly used and state-of-the-art methods.