A Volumetric Deep Learning Framework for Spinal Cord MRI Classification with Explainable AI Integration

In this work, a volumetric deep learning approach of classifying spinal cord MRI is proposed that combines 3D convolutional neural networks (CNNs) with explainable artificial intelligence (XAI). Framework attracts the essential issues of clinical AI adoption, especially the transparency and interpretability. axial and sagittal DICOM images are processed in the system based on normalization of the intensity values and spatially aligning the resulting slices, producing 3D volumes with the anatomical context intact. An adapted ResNet-18 model is used, with a correct classification rate of 91.3 percent (95 percent CI 2.1 percent). This combination of Grad-CAM and LIME XAI returns localized, validated heatmaps, which were checked with a radiologist concordance study (k = 0.78-0.83), the localized heatmaps are reliable, and the final decision of the model can be explained. The system is able to capture 3D pathological details as low as 5mm making it sensitive even in the early detection of lesions. Also, bimodal XAI algorithms performing a combination of gradient-based and perturbation techniques increase the transparency of the model, essential to clinical practice. The system has been shown to possess an 89 percent increase in terms of the diagnostic confidence score in practical clinical terms. Such a framework will provide a model of transparent and trustworthy AI-powered evaluations of spinal disorders in which early detection and interpretability are of paramount importance to facilitating treatment planning.