对比学习驱动的医学影像分割单源域泛化

Contrastive Learning-Driven Medical Image Segmentation Single-Source Domain Generalization

医学影像分割中极为常见的域偏移问题会导致深度网络应用于新目标域时遭遇性能下降。为此,本文提出了一个单源域泛化训练框架ContraSDG。该框架基于对比学习,使用单源域训练数据。通过消除深度网络对风格信息的依赖,使其专注于语义信息的学习,从而学习出鲁棒的特征表示,达到提升泛化能力的目的。实验结果表明,对于医学影像跨域分割任务,本文方法能够较大幅度地提高分割性能。

The common domain offset problem in medical image segmentation can lead to performance degradation when deep networks are applied to new target domains.This article proposes a single source domain generalization training framework ContraSDG.This framework is based on contrastive learning and uses single source domain training data.By eliminating the dependence of deep networks on style information and focusing on learning semantic information,robust feature representations can be learned,achieving the goal of improving generalization ability.The experimental results show that for cross domain segmentation tasks of medical images,our method can significantly improve segmentation performance.