Diagnostic classification of autism spectrum disorder using sMRI improves with the morphological distance-related features compared to morphological features

Abstract

In this study, we analyzed the performance of the morphological features (MF) and morphological distance-related features (MDRF) in the classification of autism spectrum disorder (ASD) and typical development (TD). Initially, we preprocessed the structural magnetic resonance images (sMRI) of ASD and TD from seven sites publicly available in the autism brain imaging data exchange (ABIDE-I and ABIDE-II) database using a standard pipeline. Further, sMRI images were parcellated into different regions using the Destrieux atlas. Moreover, MF (surface area) and MDRF were calculated from each region. We tested the performance of the MF and MDRF on each site by feeding them to classifiers such as random forest (RF), support vector machines (SVM), and multi-layer perceptron (MLP). Our results suggested that the MDRF could classify the ASD and TD better than the MF. Furthermore, RF gave a single-site average classification accuracy of 91.78% and 95.27% using MF and MDRF, respectively. We achieved the average classification accuracy of 69.08% and 82.91% between the sites using MF and MDRF, respectively. Our results suggested that the frontal lobe and right hemisphere contributed more MDRF to the machine learning model. Furthermore, in the USM site, many features were found to be from the frontal lobe (15 distance features) and frontal-parietal (11 distance features) lobe. The results suggest that the MDRF can be a valuable feature metric for classifying ASD-like neurodevelopmental disorders. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.