Domain Generalizer
Few-shot meta-learning for robust medical image segmentation across unseen domains.
Overview
Domain Generalizer is a model-agnostic meta-learning framework designed to improve the robustness of medical image segmentation when the test distribution differs from the training domains.
The method learns domain-agnostic feature representations and supports rapid few-shot adaptation to a previously unseen domain. It was evaluated on computed-tomography vertebrae segmentation across healthy and pathological datasets acquired under different conditions.