An enhanced effect-size thresholding method for the diagnosis of Autism Spectrum Disorder using resting state functional MRI

Autism Spectrum Disorders (ASD) represent a cluster of relatively common developmental conditions which require an early and accurate diagnosis for an effective remedial therapy. Resting state functional MRI (rs-fMRI) is considered an important tool to investigate the differences in functional connectivity due to ASD. In this paper, an Enhanced Effect-Size Thresholding (EEST) method is developed for extracting connectivity based features to diagnose ASD automatically from rs-fMRI. In this method, a whitening step is first used to decorrelate the Blood Oxygen Level Dependent (BOLD) signals (time-series) from the 90 representative regions of the brain based on the Automated Anatomical Labeling (AAL) template. Using these whitened time-series signals, the group-wise (ASD versus Neurotypical) differences in pairwise-connectivity are compared based on their effect-size. The connections corresponding to larger values of effect-size are alone considered for feature extraction. The z-transformed correlation co-efficients are used as features and the classification is performed using a support vector machine. The publicly available Autism Brain Imaging data Exchange (ABIDE) dataset is used to evaluate the performance of EEST and it is found that EEST can achieve better classification performance when compared to the earlier method.