Functional Connectivity Based Classification for Autism Spectrum Disorder Using Spearman’s Rank Correlation

Due to the continuous advances in brain imaging technology, an increasing number of large-scale brain research projects have been derived, such as the ABIDE Initiative. These developments have enabled us to gain an unprecedentedly detailed insight into brain activity by analyzing brain imaging data, which will reshape our understanding of brain activity and uncover biomarkers of brain disease.Over the past decade, the analysis of resting-state functional connectivity has become a trend because brain connectivity provides an effective way to understand how spatially distant brain regions interact and achieve coherent neural functions. One of the most common approaches to analyze functional connectivity is the Pearson correlation. This paper uses a new correlation method to calculate functional connectivity: Spearman’s rank correlation. We apply two conventional machine learning methods to classify autism spectrum disorder (ASD) patients from typically developing (TD) participants based on functional connectivity derived from resting-state functional magnetic resonance imaging (fMRI) data. To verify the feasibility and validity of Spearman’s rank correlation in the classification of autism, we compared the accuracy, sensitivity, and specificity of methods using functional connectivity obtained from Pearson’s correlation and Spearman’s rank correlation. Moreover, feature selection is one of the essential tasks in classification studies. We present an empirical comparison of two feature selection methods: select from model (SFM) and recursive feature elimination (RFE).