Spinal Lesion Detection in X-Ray Images via Edge Guidance and Context Aggregation

Spinal diseases typically cause serious consequences such as limited mobility and nerve damage, making accurate diagnosis clinically important for effective treatment. X-ray imaging plays a crucial role in diagnosing spinal diseases, so effectively detecting spinal lesions from X-ray images is of practical value. Considering that spinal lesions typically occur in the spine region and commonly lead to changes in the structural relationships among vertebrae, this paper proposes a novel edge-guided and context-aggregated network (EGCA-Net) for spinal lesion detection. In particular, an edge-guided spine feature enhancement module is proposed to enhance the spine features with the guidance of the spine edge maps, thus obtaining discriminative spine features. Additionally, a context-aggregated structural relationship reasoning module is designed to obtain refined features of potential lesion regions by exploring the correlation between the potential lesion and context regions. With the designed edge-guided spine feature enhancement and context-aggregated structural relationship reasoning modules, the proposed EGCA-Net effectively explores the properties of spinal lesions, thus boosting spinal lesion detection performance. Experiments on the public spine X-ray dataset have validated the effectiveness of the proposed EGCA-Net.