[1]段逸凡,肖洪兵,Rahman Md Mostafizur.基于3DSEU-Net不确定性循环焦点平均教师的半监督脑肿瘤分割[J].中国医学物理学杂志,2023,40(9):1121-1126.[doi:DOI:10.3969/j.issn.1005-202X.2023.09.010]
　DUAN Yifan,XIAO Hongbing,Rahman Md Mostafizur.Semi-supervised learning for brain tumor segmentation through 3DSEU-Net as uncertainty-aware mean teacher and cyclical focal loss[J].Chinese Journal of Medical Physics,2023,40(9):1121-1126.[doi:DOI:10.3969/j.issn.1005-202X.2023.09.010]
点击复制

基于3DSEU-Net不确定性循环焦点平均教师的半监督脑肿瘤分割()
分享到：

《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]

卷:
40卷
期数:
2023年第9期
页码:
1121-1126
栏目:
医学影像物理
出版日期:
2023-09-26

文章信息/Info

Title:
Semi-supervised learning for brain tumor segmentation through 3DSEU-Net as uncertainty-aware mean teacher and cyclical focal loss
文章编号:
1005-202X（2023）09-1121-06
作者:
段逸凡; 肖洪兵; Rahman Md Mostafizur
北京工商大学人工智能学院， 北京 100048
Author(s):
DUAN Yifan;  XIAO Hongbing;  Rahman Md Mostafizur
School of Artificial Intelligence, Beijing Technology and Business University, Beijing 100048, China
关键词:
三维卷积神经网络; 通道注意力; 半监督学习; 脑肿瘤分割; 循环焦点损失
Keywords:
Keywords: three-dimensional convolutional neural network squeeze and excitation semi-supervised learning;  brain tumor segmentation cyclical focal loss
分类号:
R318；TP183
DOI:
DOI:10.3969/j.issn.1005-202X.2023.09.010
文献标志码:
A
摘要:
准确、完整地定位和分割脑肿瘤对脑胶质瘤患者的存活率以及治疗方案的确定起着决定性作用。在三维核磁共振影像（MRI）中，生成准确的注释需要大量的专业知识和时间成本，使用少量有标签数据与大量无标签数据进行半监督学习更加符合实际的临床场景与需求。为此，本文提出一种3DSEU-Net作为半监督模型中的教师与学生网络，该网络引入注意力计算，同时结合跳跃连接，以便获取三维医学影像中更加丰富鲁棒的结构与细节特征，训练过程中，教师模型通过不确定性量化，然后指导学生模型，使学生模型学习到置信度更高的结果，在仅有少量有标签数据的情况下学习到更多的知识，以提升模型的脑肿瘤分割精度。在仅有25个有标签数据的情况下，分割精度比全监督学习提升了12.9%，最高分割精度达81.41%，优于目前可同基准复现的6种半监督方法，证明了本文方法的可行性和有效性。
Abstract:
Abstract: The accurate localization and segmentation of brain tumors greatly affects the survival rate of glioma patients and the determination of treatment schemes. Generating accurate annotations in three-dimensional (3D) magnetic resonance imaging (MRI) requires a lot of professional?nowledge and is time-consuming. The semi-supervised learning using a small amount of labeled data and a large amount of unlabeled data is more practical in clinic. Herein a 3DSEU-Net in which squeeze and excitation block is introduced and combined with skip connections is proposed as teacher and student networks in the semi-supervised model, so that the richer and more robust structural and detailed features can be extracted from 3D medical image. During training, the teacher model guides the student model by quantifying uncertainties, which makes the student model learn the results with higher degree of confidence. The proposed model is able to learn more knowledge under the condition that only a small amount of labeled data is available, thereby improving the segmentation accuracy of brain tumors. In the case of only 25 labeled data, the proposed method improves segmentation accuracy by 12.9% over fully supervised learning, and has a highest segmentation accuracy of 81.41%, outperforming 6 semi-supervised methods currently reproducible on the same benchmark. These results verify the feasibility and effectiveness of the proposed method.

相似文献/References:

[1]何兰,吴倩. 基于3D卷积神经网络的肝脏自动分割方法[J].中国医学物理学杂志,2018,35(6):680.[doi:DOI:10.3969/j.issn.1005-202X.2018.06.012]
　HE Lan,WU Qian. Automatic liver segmentation based on three-dimensional convolutional neural network[J].Chinese Journal of Medical Physics,2018,35(9):680.[doi:DOI:10.3969/j.issn.1005-202X.2018.06.012]
[2]尹梓名,董东民,陈涛.基于3D-ResNet深度影像特征的胆囊癌生存预测模型[J].中国医学物理学杂志,2022,39(7):919.[doi:DOI:10.3969/j.issn.1005-202X.2022.07.022]
　YIN Ziming,DONG Dongmin,CHEN Tao.Predictive model for survival in gallbladder cancer patients based on 3D-ResNet deep image features[J].Chinese Journal of Medical Physics,2022,39(9):919.[doi:DOI:10.3969/j.issn.1005-202X.2022.07.022]
[3]顾德,王宁,张寅斌,等.基于深度学习的三维肿瘤及器官分割[J].中国医学物理学杂志,2024,41(9):1122.[doi:DOI:10.3969/j.issn.1005-202X.2024.09.009]
　GU De,WANG Ning,ZHANG Yinbin,et al.Three-dimensional tumor and organ segmentation based on deep learning[J].Chinese Journal of Medical Physics,2024,41(9):1122.[doi:DOI:10.3969/j.issn.1005-202X.2024.09.009]

备注/Memo

备注/Memo:
【收稿日期】2023-05-06 【基金项目】北京市自然科学基金-北京市教育委员会科技计划重点联合项目（KZ202110011015） 【作者简介】段逸凡，在读研究生，研究方向：图像处理、机器学习，E-mail: 1106811434@qq.com 【通信作者】肖洪兵，博士，副教授，研究生导师，研究方向：图像处理与模式识别，E-mail: x.hb@163.com
更新日期/Last Update: 2023-09-26