福建电脑2024，Vol.40Issue(6)：1-7,7.DOI:10.16707/j.cnki.fjpc.2024.06.001
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
摘要
Abstract
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.关键词
医学影像分割/域偏移/单源域泛化Key words
Medical Image Segmentation/Domain Offset/Single Source Domain Generalization分类
信息技术与安全科学引用本文复制引用
肖榕..对比学习驱动的医学影像分割单源域泛化[J].福建电脑,2024,40(6):1-7,7.基金项目
本文得到福建省自然科学基金(No.2022J01574)资助. （No.2022J01574）
