This study presents a deep learning-based approach for the classification of cervical stenosis using MRI spine images, integrating multiple phases such as preprocessing, segmentation, feature extraction, and classification. A U-Net-based segmentation model effectively delineates key anatomical structures, including the spinal canal, intervertebral discs (IVDs), and neural foramen, improving feature extraction and classification accuracy. Furthermore, ResNet-50 is employed for feature map generation, leveraging deep hierarchical representations to extract meaningful spatial patterns from MRI slices. For classification, a Vision Transformer (ViT)-based model is utilized, taking advantage of its self-attention mechanism to capture both local and global dependencies within MRI images. Unlike conventional CNN-based models, ViT processes MRI scans as patches, enabling a more context-aware analysis of stenotic regions. The model is trained using an 80%–20% train-test split and evaluated using standard performance metrics, achieving an accuracy of 92.60%, precision of 90.16%, recall of 95.43%, and an F1-score of 91.56%. These results indicate that the ViT model outperforms traditional CNN-based classifiers in cervical stenosis detection, ensuring higher sensitivity and specificity in real-world clinical applications.