A Keypoint Transformer to Discover Spine Structure for Cobb Angle Estimation

Abstract

Due to the ambiguity and variability of vertebrae in X-ray images, accurate measurement of Cobb angles becomes challenging. Current studies are mainly focused on direct transfer learning from semantic segmentation models or keypoint estimators. We argue that such methods ignore relationships between local vertebrae shapes and the global spine structure. To address the above issue, we propose a multi-task method that ensures the consistency between semantic masks and keypoints and a keypoint Transformer to discover the spine structure. Extensive experiments on anterior-posterior (AP) X-ray images from AASCE MICCAI 2019 Challenge demonstrate the superiority of our proposed method compared to other state-of-the-art models.