Improved Classification Accuracy for Diagnosing the Early Stage of Parkinson’s Disease Using Alpha Stable Distribution

Abstract

Early diagnosis of the neurodegenerative disorder Parkinson’s disease (PD) and Scan without Evidence of Dopaminergic Deficit (SWEDD) is essential for effective patient management in neurodisorders, as both have the same clinical characteristics. The present work intends to propose an efficient method for analyzing volume rendering Single-Photon Emission Computed Tomography (SPECT) image slices, using Alpha stable distribution-based intensity normalization techniques for discriminating early PD from Healthy Control (HC) and SWEDD. The Volume rendering image (VRI) slices of early PD, HC and SWEDD are chosen from the database called Parkinson’s Progression Markers Initiative (PPMI). The alpha stable distribution technique is adapted to normalize the intensity values outside the striatum of the VRI in order to keep up the uniform intensity values throughout the database images. The shape features and image surface-related features are taken out from the VRI slices of the three different groups. The most optimized feature set is computed based on its consistency by Genetic Algorithm (GA). The computed optimized features of VRI slices show a remarkable performance in detecting the early stage of PD. The Extreme learning machine (ELM) classifier with Radial Basis Function (RBF) kernel shows a better performance accuracy of 99.12% than Support vector machine (SVM). Performance measures of the classifiers have ensured the validity of the experiments. Thus, the proposed method is advantageous to the neurologist in the early diagnosis of PD.