The Performance of Local-Learning Based Clustering Feature Selection Method on the Diagnosis of Parkinson's Disease Using Structural MRI

Abstract

The neurodegenerative diseases are modelled by the deformation of the brain neurons. In the detection of neurodegenerative diseases including Parkinson's Disease (PD), the Three-Dimensional Magnetic Resonance Imaging (3D-MRI) has been utilized, recently. In this paper, by using a Voxel-Based Morphometry (VBM) method, the morphological alterations between the Structural MRI (sMRI) data of 40PD and 40 Healthy Controls (HCs) have been determined. By using the structural alterations between the PD patients and HC and two sample t-test method, the 3D Gray Matter (GM) and White Matter (WM) tissue masks are obtained separately for two different hypotheses, t-contrast and f-contrast. The Feature Selection and Kernel Learning for Local Learning-based Clustering (LLCFS) method is used to rank the features and a Fisher criterion algorithm is utilized to determine the number of the topranked features. The selected features are classified by using two classification approaches, namely Support Vector Machines (SVM) and Naive Bayes (NB). The results indicate that the classification performances of both NB and SVM methods with f-contrast outperform that with t-contrast for all GM, WM, and the concatenation of GM and WM tissue volumes. Additionally, the classification performance of SVM is higher than that of NB for all GM, WM, and the combination of GM and WM tissues. The highest area under curve results are obtained as 75.63%, 85.00%, and 90.00% for GM, WM, and the concatenation of them, respectively.