• Kidney contours are represented as point clouds, creating a domain-invariant structural abstraction that effectively bridges different imaging modalities. • The proposed CAS model integrates structural learning from point clouds with texture features from images, enhancing segmentation through inter-modal knowledge distillation. • The PSC-UDA framework enables unsupervised 3D domain adaptation without accessing source data, improving adaptability and robustness by leveraging structural consistency. • Extensive validation shows the framework outperforms generative UDA methods and achieves state-of-the-art results in cross-site and cross-domain kidney segmentation.
Li et al. (Fri,) studied this question.