LIGHTWEIGHT REG NET-DRIVEN DEEP LEARNING FRAMEWORK FOR ENHANCED CLASSIFICATION OF NEURODEGENERATIVE DISEASES FROM MRI IMAGES

Abstract Neurodegenerative diseases significantly impact global health, necessitating early and accurate diagnosis. Current diagnostic methods combining MRI imaging and traditional machine learning algorithms often face accuracy limitations and computational inefficiencies. This research presents an efficient deep learning framework that employs an adaptive cross-guided bilateral filter for artifact removal, a lightweight Regularized Network (RegNet) with depth wise separable convolutions for multi-scale feature extraction, and ensemble machine learning techniques for accurate classification into Alzheimer's disease (AD), Parkinson's disease (PD), or Normal categories. This approach maintains high accuracy while substantially reducing computational demands. The proposed framework's efficacy is evaluated using precision, recall, F1-score, Area under the Curve (AUC), and accuracy metrics, demonstrating significant practical potential in clinical settings. Keywords: Neurodegenerative diseases, MRI, Deep learning, RegNet, Ensemble classification, Depth wise separable convolution, Artifact removal, Computational efficiency.