A Spine CT Image Segmentation Method based on an Improved U-Net

Introduction: Spine CT image segmentation is challenging due to issues such as tissue mixing, low contrast, and edge blurring, which often lead to unsatisfactory results. Method: This paper proposes the RSAFF network, which incorporates a residual self-attention module to enhance edge information extraction and an RFA_a module for feature recalibration and gradient preservation. Results: The proposed method achieved a Dice similarity coefficient of 91.52% on the test set, exceeding U-Net by 2.01% and outperforming other comparative models. Discussion: The improvement can be attributed to the enhanced ability to capture fine anatomical structures and edge details through self-attention and feature fusion mechanisms. Conclusion: The RSAFF model demonstrates significant potential for improving the accuracy of spine CT image segmentation in clinical applications.