An Ensembled Grid based Machine Learning Approach For PD Classification From MRI Images

Parkinson’s disease (PD) is a neuro-degenerative disease caused due to breakdown of brain cells in the central-part of the nervous system. As symptoms of PD appear only after 60% or more of these cells are destroyed, early detection is a tedious task. Several studies have proven that PD can be effectively diagnosed from MR-Images (MRI). This paper provides an ensembled machine learning model, Stacked-EG model, for MRI-image based PD classification. The dataset was obtained from PPMI (Parkinson’s Progression Markers Initiative) online repository which included 134 PD subjects (80 males and 54 females) and 126 healthy patients (60 males and 66 females) around 60 years old on average. Feature extraction is performed through the application of filters, including wavelet transformations and intensity-based enhancements. Feature selection is conducted using optimised XGBoost model’s feature significance scores. For classification, a Stacking Classifier which combines the strengths of XGBoost, SVM, and a Random Forest classifier was used. The model obtained an accuracy of 98.41% in classifying PD from healthy controls.