Segmentation of Vertebral X-ray Image Based on Recurrent Residual Skip Connection Structure

In recent years, the high prevalence, strong impact, and increasing occurrence among young individuals have drawn widespread attention to the spinal deformity disease. Traditional diagnostic methods for this disease suffer from drawbacks such as high error rates, long processing times, and low diagnostic efficiency. Therefore, to swiftly extract target regions like the spine and vertebrae, this paper proposes a vertebral X-ray image segmentation method based on a cyclic residual skip-connection network. Firstly, a cyclic residual skip-connection network, R2SC-Net, is introduced to perform initial segmentation of the spine and sacrum regions in the four-view images, thereby localizing the spinal regions in the images. Secondly, an algorithm is devised based on dynamic patch extraction for vertebral segmentation. This algorithm combines image morphology techniques within the target region to achieve fine-grained segmentation of individual vertebrae. Additionally, a series of image optimization processes are employed to enhance the accuracy of the predicted results.In the 228 spine image datasets collated in collaboration with the Department of Spine Surgery of the Affiliated Hospital of Qingdao University, the algorithm improved the IOU values of vertebral body segmentation in coronal, left Bending and right Bending positions by 13.32%, 8.91% and 14.27%, respectively, compared with the vertebral body segmentation algorithms of cascade positioning FCN and segmentation FCN, resulting in significant improvement in segmentation performance.