Conference Proceedings

Abstract

The aim of this paper is to study the potential use of an intelligent/automatic classification system for Parkinson's Disease (PD), using magnetic resonance images (MRI). For feature extraction of MRI Discrete Wavelet Transform has been used, followed by minimum Redundancy Maximum Relevance critrium (mRMR) for feature selection and Principal Component Analysis (PCA) for feature reduction. We then applied Support Vector Machine (SVM) for classification and Genetic Algorithms (GA) for optimization. To discover which slices and regions of the MRI are the most relevant in the brain for identification of Parkinson's Disease region, optimization was carried out using a multi-objetive genetic algorithm. The slices obtained in this optimization process were consistent with those recommended by medical experts. The methodology presented outperformed most of the research available in the bibliography, achieving accuracies of 95% in classification of subjects. This suggests that the proposed workflow and its application could help in the investigation of Parkinson's disease and aid in the research of other neurodegenerative diseases, improving their diagnostic accuracy and identifying the most relevant regions of the brain associated with each disease.