物理·技术·生物
ENGLISH ABSTRACT
多尺度融合与注意力结合的头颈部危及器官自动分割
林小惟
杨瑞杰
李霓
齐琦
作者及单位信息
·
DOI: 10.3760/cma.j.cn113030-20220128-00047
Automatic segmentation of organs at risk in head and neck carcinoma from radiation therapy using multi-scale fusion and attention based mechanisms
Lin Xiaowei
Yang Ruijie
Li Ni
Qi Qi
Authors Info & Affiliations
Lin Xiaowei
School of Computer Science and Technology, Hainan University, Haikou 570228, China
Yang Ruijie
Department of Radiation Oncology, Peking University Third Hospital, Beijing 100191, China
Li Ni
School of Mathematics and Statistics, Hainan Normal University, Haikou 571158, China
Qi Qi
School of Computer Science and Technology, Hainan University, Haikou 570228, China
·
DOI: 10.3760/cma.j.cn113030-20220128-00047
346
65
0
0
2
1
PDF下载
APP内阅读
摘要

目的开发一种多尺度融合与注意力机制结合的头颈部肿瘤放疗危及器官图像分割方法。

方法基于U-Net卷积神经网络,为增强分割模型的特征表达能力,将空间和通道注意力模块与U-Net模型相结合,提高与分割任务相关性更大的特征通道权重;在网络模型编码阶段引入本文提出的多尺度特征融合算法,补充模型下采样过程中损失的特征信息。使用戴斯相似性系数(DSC)和95%豪斯多夫距离(HD)作为不同深度学习模型之间比较的性能评估标准。

结果在医学图像计算和计算机辅助干预国际会议(MICCAI)StructSeg 2019数据集上进行头颈部22个危及器官的分割。相比于已有方法,本文提出的分割方法平均DSC提升了3%~6%,22种头颈部危及器官的分割平均DSC为78.90%,平均95%HD为6.23 mm。

结论基于多尺度融合和注意力机制的U-Net卷积神经网络对头颈部危及器官达到了更好的分割精度,有望在临床应用中提高医生的工作效率。

深度学习;卷积神经网络;危及器官图像分割;注意力机制;多尺度融合
ABSTRACT

ObjectiveTo develop a multi-scale fusion and attention mechanism based image automatic segmentation method of organs at risk (OAR) from head and neck carcinoma radiotherapy.

MethodsWe proposed a new OAR segmentation method for medical images of heads and necks based on the U-Net convolution neural network. Spatial and channel squeeze excitation (csSE) attention block were combined with the U-Net, aiming to enhance the feature expression ability. We also proposed a multi-scale block in the U-Net encoding stage to supplement characteristic information. Dice similarity coefficient (DSC) and 95% Hausdorff distance (HD) were used as evaluation criteria for deep learning performance.

ResultsThe segmentation of 22 OAR in the head and neck was performed according to the medical image computing computer assisted intervention (MICCAI) StructSeg2019 dataset. The proposed method improved the average segmentation accuracy by 3%-6% compared with existing methods. The average DSC in the segmentation of 22 OAR in the head and neck was 78.90% and the average 95%HD was 6.23 mm.

ConclusionAutomatic segmentation of OAR from the head and neck CT using multi-scale fusion and attention mechanism achieves high segmentation accuracy, which is promising for enhancing the accuracy and efficiency of radiotherapy in clinical practice.

Deep learning;Convolutional neural networks;Organs at risk image segmentation;Attention mechanisms;Multi-scale fusion
Qi Qi, Email: nc.defudabe.unaniahiqq
Li Ni, Email: nc.defudabe.unniahinil
引用本文

林小惟,杨瑞杰,李霓,等. 多尺度融合与注意力结合的头颈部危及器官自动分割[J]. 中华放射肿瘤学杂志,2023,32(04):319-324.

DOI:10.3760/cma.j.cn113030-20220128-00047

PERMISSIONS

Request permissions for this article from CCC.

评价本文
*以上评分为匿名评价
鼻咽癌发生在鼻腔的顶部和侧壁,是中国高发的恶性肿瘤之一 1。由于鼻咽癌对放射线敏感性较高,放疗是其主要治疗手段 2 , 3。通过放射线照射肿瘤杀死癌细胞,多数患者放疗后能达到根治目的 4。放疗医师需要对肿瘤周围正常、易受损伤的器官进行勾画,来确定射线照射的剂量和区域,这些器官即危及器官。目前,放疗医师通常人工勾画危及器官,该方法不仅劳动强度大、耗时长,且勾画的结果受主观因素影响较大。头颈部器官较多且集中,器官间的边界和体积差异较大,解剖关系复杂,某些器官和周围组织对比度较低,增加了头颈部危及器官分割的挑战性。
近年来,深度学习已经成为机器学习中突出的研究领域。卷积神经网络通过卷积操作提取图像特征,能有效提取图像中的信息,在医学图像分割领域得到了广泛应用 5 , 6 , 7。Ibragimov和Xing 8第一次将深度学习算法应用在了头颈部危及器官CT图像分割中,但这种方法没有考虑CT图像切片间的信息。Gao等 9使用了多个组合的卷积神经网络分割头颈部危及器官,先通过检测网络检测出器官的大致位置,再将图像放到分割网络中进行分割。Zhang等 10采用切片分类的神经网络模型对头颈部危及器官进行分割。然而,目前普遍采用的二维卷积神经网络,对医学三维图像中切片之间的上下文的空间信息学习不足。为解决上述问题,本研究基于3D U-Net卷积神经网络对头颈部危及器官分割方法做了进一步优化,提出一个新型的头颈部肿瘤放疗危及器官分割神经网络模型 11
试读结束,您可以通过登录机构账户或个人账户后获取全文阅读权限。
参考文献
[1]
Chang ET , Adami HO . The enigmatic epidemiology of nasopharyngeal carcinoma[J]. Cancer Epidemiol Biomarkers Prev, 2006,15(10):1765-1777. DOI: 10.1158/1055-9965.EPI-06-0353 .
返回引文位置Google Scholar
百度学术
万方数据
[2]
曾雷,龚晓昌,敖帆,. 基于调强放疗的鼻咽癌分期探讨[J]. 中华放射肿瘤学杂志, 2015,24(3):285-289. DOI: 10.3760/cma.j.issn.1004-4221.2015.03.014 .
返回引文位置Google Scholar
百度学术
万方数据
Zeng L , Gong XC , Ao F ,et al. Evaluation of staging system for nasopharyngeal carcinoma based on intensity- modulated radiotherapy[J]. Chin J Radiat Oncol, 2015,24(3):285-289. DOI: 10.3760/cma.j.issn.1004-4221.2015.03.014 .
Goto CitationGoogle Scholar
Baidu Scholar
Wanfang Data
[3]
林锦,韩露,林少俊,. 202例老年鼻咽癌放化疗疗效分析[J]. 中华放射肿瘤学杂志, 2013,22(6):461-464. DOI: 10.3760/cma.j.issn.1004-4221.2013.06.011 .
返回引文位置Google Scholar
百度学术
万方数据
Lin J , Han L , Lin SJ ,et al. Therapeutic effect of radiotherapy and chemotherapy in 202 elderly patients with nasopharyngeal carcinoma[J]. Chin J Radiat Oncol, 2013,22(6):461-464. DOI: 10.3760/cma.j.issn.1004-4221.2013.06.011 .
Goto CitationGoogle Scholar
Baidu Scholar
Wanfang Data
[4]
肖光莉,丘熹彬,王卫华,. 鼻咽癌调强放疗长期疗效及预后分析[J]. 中华放射肿瘤学杂志, 2012,21(6):488-491. DOI: 10.3760/cma.j.issn.1004-4221.2012.06.002 .
返回引文位置Google Scholar
百度学术
万方数据
Xiao GL , Qiu XB , Wang WH ,et al. Long-term outcome and prognostic factors of nasopharyngeal carcinoma treated by intensity modulated radiotherapy[J]. Chin J Radiat Oncol, 2012,21(6):488-491. DOI: 10.3760/cma.j.issn.1004-4221.2012.06.002 .
Goto CitationGoogle Scholar
Baidu Scholar
Wanfang Data
[5]
Han X , Hoogeman MS , Levendag PC ,et al. Atlas-based auto-segmentation of head and ne ck CT images [J]. Med Image Comput Comput Assist Interv, 2008,11(Pt 2):434-441. DOI: 10.1007/978-3-540-85990-1_52 .
返回引文位置Google Scholar
百度学术
万方数据
[6]
Shin HC , Roth HR , Gao M ,et al. Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning[J]. IEEE Trans Med Imaging, 2016,35(5):1285-1298. DOI: 10.1109/TMI.2016.2528162 .
返回引文位置Google Scholar
百度学术
万方数据
[7]
Shelhamer E , Long J , Darrell T . Fully convolutional networks for semantic segmentation[J]. IEEE Trans Pattern Anal Mach Intell, 2017,39(4):640-651. DOI: 10.1109/TPAMI.2016.2572683 .
返回引文位置Google Scholar
百度学术
万方数据
[8]
Ibragimov B , Xing L . Segmentation of organs-at-risks in head and neck CT images using convolutional neural networks[J]. Med Phys, 2017,44(2):547-557. DOI: 10.1002/mp.12045 .
返回引文位置Google Scholar
百度学术
万方数据
[9]
Gao YH , Huang R , Chen M ,et al. Focusnet: imbalanced large and small organ segmentation with an end-to-end deep neural network for head and neck CT images[C]//MICCAI. International Conference on Medical Image Computing and Computer-Assisted Intervention,Cham:Springer, 2019:829-838. DOI: 10.1007/978-3-030-32248-9_92 .
返回引文位置Google Scholar
百度学术
万方数据
[10]
Zhang S , Wang H , Tian S ,et al. A slice classification model-facilitated 3D encoder-decoder network for segmenting organs at risk in head and neck cancer[J]. J Radiat Res, 2021,62(1):94-103. DOI: 10.1093/jrr/rraa094 .
返回引文位置Google Scholar
百度学术
万方数据
[11]
Çiçek Ö , Abdulkadir A , Lienkamp S S ,et al. 3D U-Net: learning dense volumetric segmentation from sparse annotation[C]//MICCAI. International Conference on Medical Image Computing and Computer-Assisted Intervention,Cham:Springer, 2016:424-432. DOI: 10.1007/978-3-319-46723-8_49 .
返回引文位置Google Scholar
百度学术
万方数据
[12]
Yu F , Koltun V . Multi-csale context aggregation by dilated convolutions[EB/OL].( 2015-11-23)[2021-12-01]. https://arxiv.org/abs/1511.07122.
返回引文位置Google Scholar
百度学术
万方数据
[13]
Szegedy C , Liu W , Jia YQ ,et al. Going deeper with convolutions[EB/OL]. ( 2014-09-17)[2021-12-01]. https://arxiv.org/abs/1409.4842.
返回引文位置Google Scholar
百度学术
万方数据
[14]
Oktay O , cshlemper J , Folgoc L L ,et al. Attention U-Net: learning where to look for the pancreas[EB/OL]. ( 2018-04-11)[2021-12-01]. https://arxiv.org/abs/1804.03999.
返回引文位置Google Scholar
百度学术
万方数据
[15]
Roy AG , Navab N , Wachinger C . Concurrent spatial and channel 'squeeze & excitation' in fully convolutional networks [EB/OL]. ( 2018-03-07)[2021-12-01]. https://arxiv.org/abs/1803.02579.
返回引文位置Google Scholar
百度学术
万方数据
[16]
Hu J , Shen L , Albanie S ,et al. Squeeze-and-excitation networks[J]. IEEE Trans Pattern Anal Mach Intell, 2020,42(8):2011-2023. DOI: 10.1109/TPAMI.2019.2913372 .
返回引文位置Google Scholar
百度学术
万方数据
[17]
Ronneberger O , Ficsher P , Brox T . U-Net: convolutional networks for biomedical image segmentation[EB/OL]. ( 2015-05-18)[2021-12-01]. https://arxiv.org/abs/1505.04597.
返回引文位置Google Scholar
百度学术
万方数据
[18]
Kingma DP , Ba J . Adam: a method for stochastic optimization[EB/OL]. ( 2014-12-22)[2021-12-01]. https://arxiv.org/abs/1412.6980.
返回引文位置Google Scholar
百度学术
万方数据
[19]
慕光睿,杨燕平,高耀宗,. 基于多尺度三维卷积神经网络的头颈部危及器官分割方法[J]. 南方医科大学学报, 2020,40(4):491-498. DOI: 10.12122/j.issn.1673-4254.2020.04.07 .
返回引文位置Google Scholar
百度学术
万方数据
Mu GR , Yang YP , Gao YZ ,et al. Multi-scale 3D convolutional neural network-based segmentation of head and neck organs at risk[J]. J South Med Univ, 2020,40(4):491-498. DOI: 10.12122/j.issn.1673-4254.2020.04.07 .
Goto CitationGoogle Scholar
Baidu Scholar
Wanfang Data
备注信息
A

林小惟:算法设计、模型训练和论文撰写;杨瑞杰、齐琦、李霓:课题设计、问题模型的构建、算法思想论证及论文修改

B
齐琦,Email: nc.defudabe.unaniahiqq
C
李霓,Email: nc.defudabe.unniahinil
D
林小惟, 杨瑞杰, 李霓, 等. 多尺度融合与注意力结合的头颈部危及器官自动分割[J]. 中华放射肿瘤学杂志, 2023, 32(4): 319-324. DOI: 10.3760/cma.j.cn113030-20220128-00047.
E
所有作者均声明不存在利益冲突
F
国家自然科学基金 (11861030)
海南省自然科学基金 (2019RC176,621RC511)
国家重点研发计划 (2020YFE0202500)
北京市科技协调创新项目 (Z221100003522028)
评论 (0条)
注册
登录
时间排序
暂无评论,发表第一条评论抢沙发
MedAI助手(体验版)
文档即答
智问智答
机器翻译
回答内容由人工智能生成,我社无法保证其准确性和完整性,该生成内容不代表我们的态度或观点,仅供参考。
生成快照
文献快照

你好,我可以帮助您更好的了解本文,请向我提问您关注的问题。

0/2000

《中华医学会杂志社用户协议》 | 《隐私政策》

《SparkDesk 用户协议》 | 《SparkDesk 隐私政策》

网信算备340104764864601230055号 | 网信算备340104726288401230013号

技术支持:

历史对话
本文全部
还没有聊天记录
设置
模式
纯净模式沉浸模式
字号