分布感知均值教师网络的半监督医学影像分割  

Distribution-aware mean teacher networks for semi-supervised medical image segmentation

作　　者：赵小明 石培炼[1] 王丹丹 付有瑶 张石清 方江雄[2] Zhao Xiaoming;Shi Peilian;Wang Dandan;Fu Youyao;Zhang Shiqing;Fang Jiangxiong(School of Computer Science,Hangzhou Dianzi University,Hangzhou 310018,China;Instituted of Intelligent Information Processing,Taizhou University,Taizhou 318000,China)

机构地区：[1]杭州电子科技大学计算机学院,杭州310018 [2]台州学院智能信息处理研究所,台州318000

出　　处：《中国图象图形学报》2025年第2期575-588,共14页Journal of Image and Graphics

基　　金：国家自然科学基金项目(62276180,61976149)。

摘　　要：目的半监督方法旨在通过将未标记数据与标记数据的训练相结合,能够减少对标记数据的依赖,并取得较好的医学图像分割结果。然而,现有的半监督方法通常没有关注到标记数据和未标记数据之间的分布差异所带来的不利影响,尤其是在标记数据比例较低时,可能会严重影响模型的分割性能。方法提出一种采用分布感知的均值教师网络(distribution-aware mean teacher network,DAMTN)用于半监督医学影像分割。该方法利用标记数据和未标记数据的分布信息来指导模型的学习,以便在训练阶段使模型对标记和未标记数据的分割结果的分布尽可能相似。该方法采用老师—学生的网络架构,并嵌入了不同的注意力模块,以及分布感知(distribution-aware,DA)模块、完整性监督(integrity supervision,IS)模块和不确定性最小化(uncertainty minimization,UM)模块。结果在MICCAI 2018(Medical Image Computing and Computer Assisted Intervention Society)左心房分割挑战LA(left atrium)数据集和胰腺CT(computed tomography)数据集上的实验结果表明,该方法使用左心房10%标记数据时,获得的Dice系数、Jaccard指数、HD(Hausdorff distance)和ASD(average surface distance)分别为88.55%、79.62%、9.07和3.08,与基于不确定性的协同均值教师(uncertainty-guided collaborative mean teacher,UCMT)相比,Dice系数和Jaccard指数分别提高了0.42%和0.44%;而使用左心房20%标记数据时,获得的Dice系数、Jaccard指数、HD距离和ASD距离分别为90.55%、82.82%、5.78和1.77,与UCMT相比,Dice系数和Jaccard指数分别提高了0.14%和0.28%。该方法使用胰腺CT 10%标记数据时,获得的Dice系数、Jaccard指数、HD和ASD分别为70.20%、56.36%、15.64和3.57,与基于不确定性的互补一致性学习(uncertainty-guided mutual consistency learning,UG-MCL)相比,Dice系数和Jaccard指数分别提高了0.94%和1.06%;而使用胰腺CT 20%标记数据时,获得的Dice系数、Jaccard指数、HDObjective Medical image segmentation has great potential for application in clinical diagnosis.Although super⁃vised medical image segmentation models can achieve good segmentation results,they rely heavily on pixel-level annotated training data.However,acquiring labeled data is costly and can only be performed by experts.In particular,in the case of 3D medical datasets,the annotation process is more intricate and time-consuming than in 2D datasets.Semi-supervised learning is an effective solution that combines labeled and unlabeled data for improved segmentation results.However,existing semi-supervised methods do not address the performance impact caused by the distribution gap between labeled and unlabeled data,especially when the proportion of labeled data is low.Method A distribution-aware mean teacher net⁃work(DAMTN)is proposed for semi-supervised medical image segmentation to reduce the performance impact caused by the distribution gap between labeled and unlabeled data.This method utilizes the distribution information of both labeled and unlabeled data to guide the learning of the model,aiming to ensure that the segmentation results of labeled and unla⁃beled data have similar distributions during the training phase.The DAMTN adopts a teacher-student architecture,consist⁃ing of a teacher model and a student model.In addition,both the teacher model and the student model’s network architec⁃tures are based on the V-Net design,which includes an encoder and three decoders.The residual connections between the encoder and decoder have been removed.The decoders are differentiated by embedding different attention modules,which are used to introduce perturbations at the model level.These attention methods,including cross-sampling mutual attention(CMA),position attention(PA),and channel attention(CA),are employed to process high-level features.The CMA is used for inter-sample feature interaction and alignment,the PA is employed to handle spatial position information in the fea⁃ture maps,the CA is utilized

关 键 词：分布感知(DA) 均值教师 半监督 医学影像分割 注意力 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]