基于Mean Teacher的半监督腹部多器官CT影像分割方法  

Semi-supervised Abdominal Multi-organ CT Image Segmentation Method Based on Mean Teacher

作　　者：沈南燕[1] 王子彦 李静[1] 

机构地区：[1]上海大学机电工程与自动化学院,上海200444

出　　处：《工业控制计算机》2023年第1期107-108,共2页Industrial Control Computer

基　　金：跨模态信息融合的宫颈癌一体化智能筛查与辅助诊断的体系构建及应用推广,项目来源:上海市经济和信息化委员会,项目领域:医疗健康领域应用算法攻关。

摘　　要：针对基于监督学习的深度学习分割方法在应对腹部多器官CT影像分割任务中,标注成本过高、缺少高质量标记数据而导致分割效果不佳的问题,基于Mean Teacher提出一种结合多尺度预测的半监督腹部多器官分割方法。实验证明该方法可利用无标记数据实现多器官分割效果的较大提升。Aiming at the problem that the labeling cost of deep learning segmentation method based on supervised learning is too high in dealing with abdominal multi-organ CT image segmentation tasks,and the lack of high-quality labeling data leads to poor segmentation results,a semi-supervised abdominal multi-organ segmentation method combining multi-scale prediction is proposed based on Mean Teacher in this paper.Experiments show that this method can greatly improve the multi-organ segmentation effect by using unlabeled data.

关 键 词：Mean Teacher 半监督学习 多器官分割 

分 类 号：R814.42[医药卫生—影像医学与核医学] TP391.41[医药卫生—放射医学]