Parkinson's Disease Mid-Brain Assessment using MR T2 Images

The reduction of dopamine generating neurons in the brain regions known as substantia nigra (SN) is the reason for Parkinson's Disease (PD). To detect such symptom, for each subject, our algorithm only needs to analyze 3 slices around the center of a MRI DICOM volume, i.e., mid-brain area. In each slice, a window covering the SN becomes the region of interest (ROI) for further analysis. The ROIs are pre-processed by denoising and removing intensity non-uniformity. Local Binary Pattern (LBP) and Histogram Oriented Gradient (HOG) are used for feature extraction. Random Forest (RF) and Support Vector Machine (SVM) are used as classifiers with Principle Component Analysis (PCA) as feature reduction method. For evaluation, we use MRI T2 scans from the Parkinson's Progression Markers Initiative (PPMI) data set. We conducted experiments to illustrate the different classification capabilities of LBP, HOG and the fusion of these features for PD prognosis. Analysis shows that the SVM classifier with fusion feature descriptors has the most accurate classification outcome for PD assessment.