The Discriminative Power of White Matter Microstructures for Autism Diagnosis

Abstract

Abstract Magnetic resonance imaging (MRI)-based diagnosis for Autism spectrum disorder (ASD) has been a popular research topic. However, the discriminative power of a number of features derived from MRI have not been well examined yet. Here, we applied fixel-based analysis (FBA), a recent developed analytic framework for characterizing microstructure of white matter (WM) fiber bundles, to investigate the validity of WM microstructures in ASD diagnosis. Three microstructural metrics, including fiber density (FD), fiber bundle cross-section (FC), and the combination of FD and FC (FDC), were extracted from 26 high-functioning adults with ASD and 26 neurotypical (NT) individuals that were age, gender, handedness, and full-scale IQ matched. Different feature selection strategies were applied to identify high discriminative features, which were then submitted to a linear support vector machine (SVM) for classification. Both FD and FDC achieved over 73% accuracies in distinguishing ASD subjects from the controls; whereas, neither FC nor the microstructures of cortico-cortical connectivity showed discriminative ability. The performance was further improved by combining the WM microstructures with behavioral measures. Our results demonstrate the effectiveness of FD in the diagnosis of ASD.