Meta-learning guidance for robust medical image synthesis: Addressing the real-world misalignment and corruptions

作者：Jun Ho Lee, Daniel Kim, Taehun Kim, Mohammed A. Al‐masni, Yoseob Han, Dong‐Hyun Kim, Kanghyun Ryu · 发表于：Computerized Medical Imaging and Graphics · 年份：2025 · DOI：10.1016/j.compmedimag.2025.102506 · 被引用次数：2 · 研究领域：Generative Adversarial Networks and Image Synthesis、AI in cancer detection、Medical Image Segmentation Techniques

Deep learning-based image synthesis for medical imaging is currently an active research topic with various clinically relevant applications. Recently, methods allowing training with misaligned data have started to emerge, yet current solution lack robustness and cannot handle other corruptions in the dataset. In this work, we propose a solution to this problem for training synthesis network for datasets affected by mis-registration, artifacts, and deformations. Our proposed method consists of three key innovations: meta-learning inspired re-weighting scheme to directly decrease the influence of corrupted instances in a mini-batch by assigning lower weights in the loss function, non-local feature-based loss function, and joint training of image synthesis network together with spatial transformer (STN)-based registration networks with specially designed regularization. Efficacy of our method is validated in a controlled synthetic scenario, as well as public dataset with such corruptions. This work introduces a new framework that may be applicable to challenging scenarios and other more difficult datasets. • Medical images from different modalities are typically not aligned. • Employing meta-learning re-weighting reduces corrupted instance influence. • Addressing mis-registration, artifacts, and deformations with a meta-learning. • Validated the proposed method in synthetic scenarios and diverse datasets. • Our method enhances accuracy and robustness in medical image synthesis.