Title : A Statistical Method to Classify Parkinson’s Patterns in the Human Brain
Abstract:
Parkinson’s is a neurodegenerative progressive disease it affects human brain functionality as the disease progresses. It mostly affects movement, slows down the functionality of motor areas, and increases the difficulty in performing any task for the patients. Resting state functional Magnetic Resonance Imaging (fMRI) is the best way to distinguish changed patterns in human the brain. However, in Parkinson’s case symptoms of disease worsen over time so, Resting State fMRI scans of 11 males and 7 females are taken over 3 years. Correlation analysis showed that functional connectivity in regions of interest (Left and Right Cerebellum, right middle temporal gyrus, right middle Occipital, Left Parietal Cortex, Primary Motor Cortex, and Sensory Cortex) decrease as the disease progresses. Mediation analysis shows that year 2 of disease is more disturbed than year 1 and due to disturbance in year 2 patterns in year 3 are more affected. Ridge regression removes multi-collinearity from data and gives reliable estimates by using the regularized parameters. Different machine learning techniques like Linear Support Vector Machine, RBF Support Vector Machine, Naïve Bayes, Random Forest, Ada Boost, Neural Network, Decision Tree, Quadratic Discriminant Analysis, Gaussian Process, and Nearest Neighbours are used to classify combinations of 3 years’ data. Amon all the techniques it is concluded that Ridge regression gives the best results for the classification of Parkinson’s pattern in the human brain.