Spinal Scoliosis X-Ray Image Segmentation Network Based on Dimension Attention U-Net

Scoliosis, a common spinal deformity, significantly affects patients' posture and motor function. Traditional diagnosis of scoliosis relies on X-ray imaging to examine vertebral details and manually measure the Cobb angle to assess the degree of spinal curvature. This process is time-consuming and open to subjective influence from medical professionals. To improve diagnostic efficiency and accuracy, we propose a scoliosis X-ray image segmentation method based on a dimensional attention U-Net. This approach enables high-precision vertebral segmentation by incorporating a multi-scale convolution module that extracts complex spinal texture features across multiple spatial scales. Additionally, a global-local dimensional attention module is designed to enhance the capture of detailed vertebral information. The method also includes a feature fusion module that efficiently integrates features from different levels, strengthening the understanding of the overall spinal structure. Experimental results demonstrate that the proposed segmentation method not only enhances segmentation accuracy but also significantly improves vertebrae segmentation in spinal X-ray images, providing reliable technical support for the intelligent diagnosis of scoliosis.