Background
Existing deep learning-based image registration models are primarily designed for single organ registration. In addition, many of them are customized to perform best on a single specific organ-of-interest only, which limits its reproducibility, accessibility and generalizability in such a multidisciplinary research era.
Multi-organ registration may address these challenges but introduce additional complexities. It requires the simultaneous optimization of multiple deformation fields rather than focusing on a single one. Additionally, the deformation fields for different organs can vary significantly in terms of size, shape (e.g., pancreas vs. liver), and location (e.g., pancreas vs. lung), which further complicates the registration process.
In this study, we will develop TotalRegistrator, which is a general-purpose deep learning-based multi-organ whole body CT image registration platform. This platform is expected to become a handy tool in both research and clinical practice so that users no longer need to look for different registration methods for different organ-of-interest in their study
