Central canal spinal stenosis is a common form of degenerative spine disorder that leads to pain, numbness, and reduced mobility, particularly in elderly individuals. However, existing methods often suffer from lower sensitivity and specificity, leading to potential misclassification. Additionally, several approaches rely on limited contextual or structural features, making them ineffective in detecting minor or complex stenosis patterns. To overcome these challenges, a novel deep learning-based DESCOP approach has been proposed for spinal stenosis detection using T2-weighted Mid-sagittal lumbar MRI images. The collected images are pre-processed using a median filter to remove noise while preserving edges. Sobel edge detection is used to determine the boundaries of vertebrae followed by Multi RoI (multiple Regions of Interest) segmentation for precise localization of spinal structures. The proposed Spinal Cord Projection Descriptor integrated MobileNet (SCPDM-Net) captures both handcrafted texture patterns and deep spatial features. Fully connected Layer (FCL) is used to classify Stenosis and Non-Stenosis. The effectiveness of the DESCOP approach is assessed using parameters including specificity, recall, accuracy, and precision. The DESCOP approach achieves an overall accuracy of 99.15% for spinal stenosis detection. The proposed model improves the total accuracy by 7.30%, 5.13%, and 10.49% better than Faster R-CNN, deep CNN-based automated system, and CADx, respectively.
Kalaivani et al. (Mon,) studied this question.