Contrastive learning based method for X-ray and CT registration under surgical equipment occlusion

Xiyuan Wang, Zhancheng Zhang, Shaokang Xu, Xiaoqing Luo, Baocheng Zhang, Xiaojun Wu

Abstract: Highlights•Developed a novel contrastive learning method paired with a transformer-based network, ensuring high registration success even in occluded scenarios.•Introduced unique X-ray styles in training data via special data augmentation, enhancing model adaptability to different occlusions.•Proposed a refined RoI fine-tuning approach for efficient similarity computation in occluded states, improving accuracy to meet intraoperative demands.