[1]陈洪涛,郑芳,高艳,等.基于边缘强化的Unet-TIC模型对前列腺自动勾画研究[J].中国医学物理学杂志,2022,39(6):719-725.[doi:DOI:10.3969/j.issn.1005-202X.2022.06.011]
　CHEN Hongtao,ZHENG Fang,GAO Yan,et al.Automatic prostate segmentation with boundary-enhanced Unet-TIC model[J].Chinese Journal of Medical Physics,2022,39(6):719-725.[doi:DOI:10.3969/j.issn.1005-202X.2022.06.011]
点击复制

基于边缘强化的Unet-TIC模型对前列腺自动勾画研究()
分享到：

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

卷:
39卷
期数:
2022年第6期
页码:
719-725
栏目:
医学影像物理
出版日期:
2022-06-27

文章信息/Info

Title:
Automatic prostate segmentation with boundary-enhanced Unet-TIC model
文章编号:
1005-202X（2022）06-0719-07
作者:
陈洪涛; 郑芳; 高艳; 史亚滨; 邓小年; 钟鹤立
深圳市人民医院（暨南大学第二临床医学院，南方科技大学第一附属医院）肿瘤放疗科， 广东 深圳 518020
Author(s):
CHEN Hongtao;  ZHENG Fang;  GAO Yan;  SHI Yabin;  DENG Xiaonian;  ZHONG Heli
Department of Radiation Oncology, Shenzhen Peoples Hospital (the Second Clinical Medical College, Jinan University the First Affiliated Hospital, Southern University of Science and Technology), Shenzhen 518020, China
关键词:
Unet; Unet-TIC; CED; 边缘强化; 自动分割; Dice系数
Keywords:
Keywords: Unet Unet-TIC coherence-enhancing diffusion boundary enhancement automatic segmentation Dice similarity coefficient
分类号:
R318
DOI:
DOI:10.3969/j.issn.1005-202X.2022.06.011
文献标志码:
A
摘要:
目的：通过图像边缘强化以及改进Unet深度学习网络结构提高前列腺自动分割的准确性。方法：随机选取我院50例检查者前列腺核磁扫描图像，以及MICCAI Grand Challenge数据库50例前列腺图像，其中81例作为训练集，10例作为验证集，9例作为测试集。通过旋转不变性相干增强扩散滤波方法（CED-ORI）对图像进行边缘强化，建立双输入收缩路径结构的Unet-Two Input Channel （Unet-TIC）并行提取原始和CED-ORI图像特征，共享一条扩张路径，并通过跳跃连接突出CED-ORI边缘强化的有效特征，获取更多信息增加上采样分辨率。采用Accuracy、Mean DSC、Median DSC、ASD、MSD、RVD 6个指标对Unet、Unet-c和Unet-TIC 3种方法进行评估。结果：Unet-c和Unet-TIC评估指标表现均明显优于Unet，其中表现最好的Unet-TIC相较于Unet，Accuracy提高1.87%，Mean DSC提高1.81%，Median DSC提高1.21%，ASD降低0.32 mm，MSD降低1.63 mm，RVD降低4.64%。直观勾画方面Unet-TIC比Unet更加精准，更能捕捉到复杂的前列腺形状的变换，尤其是辨别混淆性、类似性边界区域。结论：相较于Unet，Unet-TIC在影像器官分割与勾画方面具有更优的表现。
Abstract:
Objective To improve the accuracy of automatic prostate segmentation with image boundary enhancement method and optimized Unet model. Methods The prostate scans of 50 patients in Shenzhen Peoples Hospital and 50 cases from MICCAI Grand Challenge database were analyzed in this study, with 81 cases for training set, 10 cases for validation set, 9 cases for test set. The coherence-enhancing diffusion filtering with optimized rotation invariance (CED-ORI) was used to enhance the image boundary. Unet-two input channel (Unet-TIC) was established, with two input contraction paths for the parallel capture of the features from original and CED-ORI images, sharing one expansion path, and it could highlight features contributed to CED-ORI boundary enhancement through concatenation layers, thereby capturing more multi-level information for enhancing upsampling resolution. The performances of Unet, Unet-c and Unet-TIC were evaluated by accuracy, mean DSC, median DSC, ASD, MSD and RVD. Results Both Unet-c and Unet-TIC had better segmentation performances than Unet, and Unet-TIC which had the best performance was superior to Unet in all 6 evaluation indexes. Compared with those of Unet, the accuracy, mean DSC and median DSC of Unet-TIC were improved by 1.87%, 1.81% and 1.21%, respectively, and ASD, MSD and RVD were decreased by 0.32 mm, 1.63 mm and 4.64%, respectively. Moreover, Unet-TIC was more accurate in visual delineation than Unet, and it was able to capture complex shape changes of the prostate, especially identifying confusing and similar boundary areas. Conclusion Unet-TIC is advantageous over Unet in organs segmentation and delineation.

相似文献/References:

[1]浦秀丽,刘翔,汤显,等.基于边缘监督的肝部超声图像包膜分割网络[J].中国医学物理学杂志,2022,39(10):1255.[doi:DOI:10.3969/j.issn.1005-202X.2022.10.013]
　PU Xiuli,LIU Xiang,TANG Xian,et al.Liver capsule segmentation in ultrasound image using edge supervision based network[J].Chinese Journal of Medical Physics,2022,39(6):1255.[doi:DOI:10.3969/j.issn.1005-202X.2022.10.013]
[2]魏坤,沈记全,赵艳梅.MAUNet：用于皮肤病变分割的轻量级模型[J].中国医学物理学杂志,2023,40(5):555.[doi:DOI:10.3969/j.issn.1005-202X.2023.05.006]
　WEI Kun,SHEN Jiquan,ZHAO Yanmei.MAUNet: a lightweight model for skin lesion segmentation[J].Chinese Journal of Medical Physics,2023,40(6):555.[doi:DOI:10.3969/j.issn.1005-202X.2023.05.006]

备注/Memo

备注/Memo:
【收稿日期】2021-12-25 【基金项目】广东省医学科研基金（B2021395） 【作者简介】陈洪涛，硕士研究生，研究方向：图像处理、深度学习，E-mail: taohongchen@hotmail.com 【通信作者】钟鹤立，高级工程师，研究方向：肿瘤放射物理，E-mail: zhongheli@tom.com
更新日期/Last Update: 2022-06-27