Identifying Neural Patterns and Biomarkers of ASD through Multi-phase Resting-State Functional MRI Analysis

Autism Spectrum Disorder (ASD) is a developmental disorder impacting social interaction and behavior. The current diagnosis process is based on subjective developmental and behavioral observations. As prevalent as ASD is, studying the causes of ASD, a neurological disorder, from a biological perspective presents significant challenges due to complex relations between different regions of the brain. We therefore aimed to identify and understand the neural patterns and biomarkers of ASD through a novel multiphase approach based on fMRI data analysis. We propose a three-phase approach that partitions analysis into subtasks, exploring how attributes of fMRI scans, regions of interest, and functional connectivity are uniquely associated with ASD. In the first phase, categorical analysis, we measured classification accuracy, sensitivity, and specificity variation across six parcellation atlases, 17 data collection sites, and five subject demographic categories. In the second phase, ensemble concatenation and ensemble voting techniques were used to gain additional insights on the factors that affected classification performance. The results of the first two phases revealed that although different atlases lead to similar classification decisions, each atlas formulates these decisions based on different functional or anatomical information. By examining functional connectivity differences between subjects with and without ASD in the third phase, we developed a novel method for ranking functional connections by relevance to ASD. The findings from this study contribute to a better understanding of the neural patterns in the regions and connections associated with ASD, providing insights to drive new solutions in diagnosing and providing development and treatment plans for affected individuals.