Lumbar spine disorder detection using segmentation and IBN-MGNet classification with enhanced feature extraction

Back pain affects millions worldwide, with lumbar spine disorders (LSDs) being a major contributing factor due to the spine's critical role in supporting body weight. This study aims to improve the detection of LSDs using advanced deep learning techniques. The proposed framework introduces a novel pipeline that begins with bilateral filtering for image preprocessing, followed by segmentation using a Proposed Residual Block with Patch-based RESU-NET (PRB-PRESU-NET). Feature extraction integrates Gaussian Filter with Scharr Operator-based Gradient Local Ternary Patterns (GSO-GLTP), MT, and deep features, where the Scharr operator enhances edge detection compared to conventional Sobel filters. Finally, classification is performed using the Improved Bottle Neck-based Modified GhostNet (IBN-MGNet), which achieves a high accuracy of 0.9312, outperforming existing methods. The innovation lies in the integration of refined segmentation, feature extraction, and classification strategies, resulting in a reliable and precise system for lumbar spine disorder detection.