A Space Aware Temporal Convolutional Neural Networks Approach for Enhanced Diagnosis of Multiple Sclerosis Brain Atrophy using MR images

The autoimmune condition known as multiple sclerosis (MS) leads to inflammation and demyelination, resulting in a gradual deterioration of the central nervous system. A significant sign of the advancement of MS is brain atrophy, especially in the regions of gray matter (GM) and white matter (WM). Utilizing Tl-weighted MRI slices along with patient-specific information such as age, EDSS score, and the onset of the disease, this research introduces a new deep learning framework called MSTCNN (Multi-Anchor Space-Aware Temporal Convolutional Neural Network), designed to classify brain atrophy associated with multiple sclerosis. The architecture of the model combines temporal convolution, spatial attention mechanisms, and metadata fusion to enhance prediction accuracy and interpretability in a clinical setting. Our model outperformed well-known CNN designs like ResNet18, VGG16, MobileNetV3 and InceptionNet, achieving a maximum classification accuracy of 91.8% when evaluated on a publicly available multiple sclerosis brain MRI dataset with expert-labeled images and clinical metadata. These results demonstrate how promising hybrid attention-based frameworks are for offering reliable and clinically meaningful assessments of neurodegenerative changes in MS patients.