Predictive Value of Different rs-fMRI Parameters in Wilson Disease based on SVM

Abstract

Structural and functional abnormalities have been reported in Wilson's disease (WD) patients. However, the discriminative power of combining multiple features from different analysis techniques has not been systematically investigated. We propose a novel classification technique that precisely distinguishes individuals with WD from healthy individuals. This method combines two different features extracted from resting-state functional magnetic resonance images using Regional Homogeneity (ReHo) and Amplitude of Low-Frequency Fluctuations (ALFF). We employed 32 WD patients and 26 age- and gender-matched healthy controls (HC). Three-dimensional T1 and resting-state functional magnetic resonance imaging (rs-fMRI) data were acquired from all participants and used two-sample t-tests to compare the ReHo values and ALFF values between groups. Subsequently, the support vector machine (SVM) classifier with RBF kernel was configured to evaluate the effect of abnormal rs-fMRI parameters in the differentiation of WD patients from HC. The experimental results showed that ALFF was elevated in the pallidum and thalamus and decreased in the orbital part of the inferior frontal gyrus, cingulum, and medial frontal gyrus in WD patients. At the same time, WD patients had increased ReHo in the inferior frontal gyrus, postcentral gyrus, and insula, and decreased in the superior frontal gyrus. Using leave-one-out cross-validation, the combined feature (ReHo + ALFF) yielded good classification result with 91.6% accuracy, 91.07% sensitivity, 86.93% specificity. The results of this paper show that WD patients have ReHo and ALFF abnormalities and that WD patients can be accurately identified when combining ReHo and ALFF features.