Characterization and classification Of high functioning and low Functioning autism using fmri and Deep belief networks

Abstract

Autism is a neuro-developmental disorder that includes wide range of presentations such as impaired socializing skills, language processing capabilities, repetitive and restricted behavior, poor eye contact, inclined with specific interests and so on. The spectrum of autism is mainly categorized into Low Functioning Autism (LFA) and High Functioning Autism (HFA). LFAs are the severely affected individuals who have specific traits such as poor language processing capabilities, low IQ levels, inattentiveness, etc while HFAs show extreme capabilities in specific tasks such as language processing, puzzle solving etc., but lack with a few normal skill sets. Though FAs and HFAs differ outwardly which makes them functionally different, there exists overlaps in their brain behavior that remains a stiff challenge in distinguishing both the groups. Therefore, exploring the functional nterconnections between different brain regions can highlight subtle differences that can classify the behaviorally overlapping autism spectrum. Functional Magnetic Resonance Imaging (fMRI) has been effective in imaging and capturing the neuronal correlations of the brain during rest and while performing various tasks. In this work, fMRI inputs were acquired from Autism Brain Image Data Exchange (ABIDE) for one hundred and four LFA and seventy HFA participants. Task based fMRI has also been acquired from ABIDE with a specific experimental protocol designed when the subjects ere involved in a language processing task. Artifacts were removed in the fMRI inputs using pre-processing techniques such as slice time correction, realignment and re-slicing, co-registration and normalization newline