Rest-fMRI Based Comparison Study between Autism Spectrum Disorder and Typically Control Using Graph Frequency Bands

Abstract

Graph signal processing is a subset of signal processing enabling the analysis of functional magnetic resonance imaging (fMRI) data in brain topological domain. One of the most important and highly interested tool of GSP is graph Fourier transform (GFT) by which brain signals can be analyzed in different graph frequency bands. In this paper, the resting-state fMRI (rfMRI) data is analyzed using GFT tool in order to discover new knowledge about the autism spectrum disorder (ASD) and find features discriminating between ASD and typically control (TC) subjects. For ASD group, the signal concentration in both low and high frequency bands is decreased by increasing the age in most of the brain well-known networks. The ASD in comparison to TC shows less intention for changing the signal concentration level when the level is very low or very high. In graph low frequency band, increasing the age is along with increasing the segregation and integration of brain ROIs respectively for ASD and TC. Also, in this band, the brain ROIs integration of ASD is more than TC. By increasing the age, the auditory network of ASD subjects shows increasing segregation and integration in graph low and high frequency bands, respectively. The reduced segregation of default mode network in ASD is happened in graph middle and higher frequency bands. The functional connectivity analysis between low and high frequency signals shows that some of the high frequency ROIs have connections with all low frequency ROIs so that the most of these connections are dramatically and significantly different between ASD and TC. By analyzing the local vertex frequency spectrum (LVFS) of ASD and TC at different states, it is seen these groups show contradictory behaviors with respect to each other in brain default mode network in two states. The results of different scenarios at different graph frequency bands demonstrate that using functional and structural data together can provide powerful tool for recognizing the ASD or even other brain disorders.