Deep Learning-based Parkinson disease Classification using PET Scan Imaging Data

Abstract

Parkinson's disease (PSD) is a neurodegenerative disease responsible for damaging the nerve cells inside the human brain. It is classically associated with a loss of dopaminergic neurons (DNs) inside the human brain. DNs can communicate with other nerve cells to generate smooth cooperation, but their insufficiency affects the motor and non-motor symptoms. Earlier, the PSD was recognized via manual examination of its symptoms. Researchers across the globe have given diverse automated solutions to recognize the PSD. Most existing solutions used the standard MRI and SPECT datasets for PSD recognition and less emphasis on the PET scan dataset. Existing PET scan dataset based solutions using machine learning techniques such as linear regression and SVM, which requires manual feature extraction. Motivated from these, we proposed a VGG16-based convolutional neural network (CNN) system to recognize the PSD. It automatically extracts features from the PET scan image dataset, which is collected from the PPMI source. The performance of the proposed system is evaluated using specificity, accuracy, sensitivity, and precision, which is achieved as 97.5%, 84.6%, 71.6%, and 96.7%, respectively.