SwinDiff_ConSeg: Automatic Segmentation of Ureters in Consecutive CT Urography Based on Swin Transformer and Diffusion Model

CT urography (CTU) is a widely utilized imaging technique for comprehensively examining the entire urinary system to diagnose causes of blood in the urine, known as hematuria. However, the assessment can be time-consuming as the urinary tract spans the entirety of the abdomen and pelvis and components of the urinary tract are often difficult to visualize. In particular, the ureters are very small and identifying a stone or tumor within the ureters often proves difficult for radiologists. To aid in the detection of pathology, we propose a method for segmenting the ureters. Our novel method uses consecutive slices in CTU examinations. This approach leverages both the Swin Transformer and the Diffusion Model. The multi-scaled features of the Swin Transformer aid in extracting different frequency information, and the multiple outputs of the Diffusion Model simulate the annotations provided by radiologists. The results of our experiments indicate that our method delivers exceptional segmentation outcomes, surpassing the performance of diffusion model-based approaches that utilize a single slice input.