Multiple sclerosis (MS) is an autoimmune disorder characterized by damage to the central nervous system (CNS), leading to neuronal degeneration and affecting over 2.8 million individuals globally. Early and accurate diagnosis of MS is critical, given its significant social and economic consequences. Magnetic resonance imaging (MRI) remains the gold standard for MS diagnosis and monitoring. This study introduces a novel Content-Based Medical Image Retrieval (CBMIR) framework that leverages a newly designed Convolutional Autoencoder (CAE) model to improve the diagnostic evaluation of MS-related MRI scans. The proposed system extracts latent features from query and reference MRI images using the CAE. Extensive ablation studies involving nine distance metrics and diverse feature space dimensions identify 64 as the optimal latent feature size and validate the Mahalanobis distance as the superior similarity measure. Evaluated on four publicly available MS MRI datasets, the framework achieves Mean Average Precision (MAP) scores of 91.23%, 98.68%, 99.88%, and 99.69%, respectively, demonstrating enhanced diagnostic accuracy. The system also outperforms existing similar CBMIR frameworks for other diseases in MAP scores and generalizes effectively without requiring extensive preprocessing or segmentation. The primary contribution of this work is the development of a CAE-driven CBMIR system optimized for MS diagnosis, achieving state-of-the-art MAP performance while maintaining an average retrieval latency of 780 ms outperforming the compared systems. |
