DUAL-PATH DEEP LEARNING FRAMEWORK WITH RCNN FOR AUTOMATED LUMBAR DISC HERNIATION DETECTION AND CLASSIFICATION

Abstract

Accurate classification of lumbar spine disorders using magnetic resonance imaging (MRI) remains a challenging task due to complex anatomical structures and variability in image quality. This paper proposes a dual-path deep learning framework integrated with a region-based convolutional neural network (R-CNN) for automated detection and classification of lumbar disc conditions. A pre-trained ResNet50 model is employed for robust feature extraction, followed by a dual-path architecture that captures both global structural features and local region-specific details. Additionally, a Faster R-CNN module is incorporated to localize regions of interest associated with abnormalities, improving interpretability and diagnostic reliability. The extracted features are fused and optimized using a hybrid loss function combining cross-entropy loss and cosine embedding loss to enhance classification performance and feature discrimination. The proposed model classifies MRI images into three categories: Herniated Disc (HD), No Stenosis (NS), and Thecal Sac (TS). Experimental results demonstrate that the model achieves an overall accuracy of 98.58%, along with high precision, recall, and F1-score. The findings indicate that the proposed framework is efficient, robust, and suitable for clinical decision support systems.