Graph Fourier transform of fMRI temporal signals based on an averaged structural connectome for the classification of neuroimaging

• Published in: Artificial intelligence in medicine Vol. 106; p. 101870
• Main Authors: Brahim, Abdelbasset, Farrugia, Nicolas
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2020; Elsevier
• Subjects: Artificial Intelligence; Classification; Computer Science; Graph signal processing; Machine learning; Neuroimaging; Resting-state analysis; Neuroimaging; Graph signal processing; Resting-state analysis; Machine learning; Classification; Graph Signal Processing; Machine Learning; Resting-State Analysis
• ISSN: 0933-3657; 1873-2860; 1873-2860
• DOI: 10.1016/j.artmed.2020.101870

•A novel modelling time series approach, which is applied on rs-fMRI brain imaging, is presented.•The proposed approach restores informative features related to neuro-psychiatric disease, such as Autism Spectrum Disorder, as exemplified by statistically robust gains in classification metrics when compared to other feature extraction methods.•The proposed approach decreases the amount of data needed to store patient imaging data history.•The proposed analysis method is validated on a world-wide multi-site database (ABIDE) in which different methods of imaging acquisition were used. Graph signal processing (GSP) is a framework that enables the generalization of signal processing to multivariate signals described on graphs. In this paper, we present an approach based on Graph Fourier Transform (GFT) and machine learning for the analysis of resting-state functional magnetic resonance imaging (rs-fMRI). For each subject, we use rs-fMRI time series to compute several descriptive statistics in regions of interest (ROI). Next, these measures are considered as signals on an averaged structural graph built using tractography of the white matter of the brain, defined using the same ROI. GFT of these signals is computed using the structural graph as a support, and the obtained feature vectors are subsequently benchmarked in a supervised learning setting. Further analysis suggests that GFT using structural connectivity as a graph and the standard deviation of fMRI time series as signals leads to more accurate supervised classification using a world-wide multi-site database known as ABIDE (Autism Brain Imaging Data Exchange) when compared to several other statistical metrics. Moreover, the proposed approach outperforms several approaches, based on using functional connectomes or complex functional network measures as features for classification.