PDCNet: Parkinson’s Disease Classification Network using MRI Scans

Abstract

Parkinson's Disease (PD) is a neurodegenerative condition in which timely and precise diagnosis is essential for optimal care. A Convolutional Neural Networks (CNNs) architecture, PDCNet is designed to classify PD from Magnetic Resonance Imaging (MRI) scans sourced from the NTUA Parkinson dataset. The PDCNet has three convolution layers with different numbers of filters and two fully connected layers with efficient Gradient-weighted Class Activation Mapping (Grad-CAM) to improve generalization and mitigate overfitting. The proposed PDCNet uses different dropout ratios which provides the benefits of regularization without compromising predictive performance. The proposed PDCNet has a remarkable classification accuracy of 98.27%, with a sensitivity of 97.69% and specificity of 98.85% when using 10000 images of NTUA dataset for each class (normal and PD). The experimental results emphasize the capability of PDCNet with other deep learning architectures such as VGG, ResNet, and deep belief networks for non-invasive PD classification using MRI data. The study's findings indicate that integrating AI-driven diagnostics into clinical workflows may enhance early identification andKattankulathur individualised therapy planning for individuals with PD.