Abstract
Multiple Sclerosis (MS) is a chronic autoimmune disease affecting the central nervous system, characterized by demyelination and resulting in a range of neurological impairments. While MRI is the primary imaging modality for diagnosis, manual interpretation is time-consuming and subjective. This study presents an automated deep learning approach to classify brain MRI scans for MS detection, leveraging a VGG16-based Convolutional Neural Network (CNN) enhanced via transfer learning. The model classifies images into four categories: Control-Axial, Control-Sagittal, MS-Axial, and MS-Sagittal. We evaluated the model under two scenarios: using an original dataset of 3,427 MRI images and an augmented dataset of 6,000 images. The model trained on the augmented dataset achieved superior performance with 98.09% test accuracy and 99.87% AUC, compared to 94.44% and 99.13% respectively without augmentation. Precision and recall across all classes exceeded 0.93, and ROC curves showed near-perfect separability, particularly improving distinction between MS and control in axial views. These findings suggest that deep learning, especially when paired with data augmentation, offers a reliable and efficient tool for MS detection. Future work will explore multimodal