Hallucinated Style Distillation for Single Domain Generalization in Medical Image Segmentation

Single domain generalization (single-DG) for medical image segmentation aims to learn a style-invariant representation, which can be generalized to a variety unseen target domains, with the data from a single source. However, due to the limitation of sample diversity in the single source domain, the robustness of generalized features yielded by existing methods is still unsatisfactory. In this paper, we introduce a novel single-DG framework, namely Hallucinated Style Distillation (HSD), to generate style-invariant features with consistent contents under style variations within an expanded representation space. Specifically, our HSD firstly enhances the style diversity of the single source domain via hallucinating the samples with random channel statistics. Given that out-of-distribution input impacts both the activation value statistics and activated locations, we further propose a decorrelated representation expansion method to indirectly simulate the latter scenario by broadening the representation space. Finally, a hallucinated cross-style distillation paradigm is proposed to distill the style-invariant knowledge between the original and style-hallucinated features, thereby promoting the extraction of structural information. Extensive experiments on two standard domain generalized medical image segmentation datasets show the superior performance of our HSD.