RUnT: A Network Combining Residual U-Net and Transformer for Vertebral Edge Feature Fusion Constrained Spine CT Image Segmentation

Scoliosis, spinal deformity and vertebral spondylolisthesis are disorders with high incidence, which seriously affect people’s lives health. CT is an important medical tool for the detection diagnosis of provides a large amount pathologically valid information in various clinical practices such as spine pathology assessment computer-assisted surgical interventions. As presents long span, complex shape biological curve multi-stage similarity sagittal plane images. Therefore, fast accurate segmentation technology has become research direction computer-aided diagnosis. We proposed RUnT network based on combination residual U-Net feature extraction Vision Transformer structure efficient automatic multiple vertebrae spine. The deep features first extracted using to prevent gradient diffusion while improving accuracy contour segmentation. Then multi-scale maps by containing rich superficial input edge module. designed refine boundaries ensure consistency each vertebra combining operations deconvolution convolution three different scales features.Finally, global module combined local achieve blending location through self-attentive map volume. By mixing semantic features, confusion arising from between when decoder extracts reduced. model this paper experimented CTSpine1K VerSe 20 public datasets. results show that obtains state-of-the-art performance average DSC scores 88.4% 81.5% 20, respectively, reducing distance HD95 4.86 3.88.