[1]曾安△,吴春彪,徐小维,等.基于多尺度集成的冠状动脉分割模型*[J].生物医学工程研究,2022,03:239-247.
ZENG An,WU Chunbiao,XU Xiaowei,et al.Coronary artery segmentation model based on multiscale intergration[J].Journal of Biomedical Engineering Research,2022,03:239-247.
点击复制
《生物医学工程研究》[ISSN:1006-6977/CN:61-1281/TN]
- 期数:
-
2022年03期
- 页码:
-
239-247
- 栏目:
-
- 出版日期:
-
2022-09-25
文章信息/Info
- Title:
-
Coronary artery segmentation model based on multiscale intergration
- 文章编号:
-
1672-6278 (2022)03-0239-06
- 作者:
-
曾安1△; 吴春彪1; 徐小维2; Najeeb Ullah 3
-
1.广东工业大学 计算机学院,广州 510006;2.广东省人民医院,广州 510080;3.马尔丹工程技术大学,马尔丹 23200
- Author(s):
-
ZENG An1; WU Chunbiao1 ; XU Xiaowei2 ; Najeeb Ullah 3
-
1.School of Computers, Guangdong University of Technology, Guangzhou 510006,China;2.Guangdong Provincial People’s Hospital, Guangzhou 510080;3.University of Engineering and Technology Mardan, Mardan 23200, Pakistan
-
- 关键词:
-
全卷积神经网络; U-Net; U-Net++; 体素; 图像分割; 先验提取; 数学形态学
- Keywords:
-
Fully convolutional neural networks; U-Net; U-Net++; Voxels; Image segmentation; Priori extraction; Mathematical morphology
- 分类号:
-
R318;TP399
- DOI:
-
-
- 文献标识码:
-
A
- 摘要:
-
针对传统冠状动脉分割中需要人为干预且效率低,以及现有深度学习分割方法准确率低的问题,本研究提出一种基于多尺度集成的分割模型。该模型设计了一种新的由粗到细的分割框架,通过结合全尺度的粗分割与局部多尺度的细分割,进一步提升分割的准确率。实验结果表明,在Dice相似性系数上可达到82.96%,优于其他常规的深度学习方法。该模型也为其他管状器官的分割提供了新的思路。
- Abstract:
-
To address the problem that human intervention and low efficiency in traditional segmentation methods and improve accuracy in deep learning segmentation methods, we proposed a coronary artery segmentation model based on multi-scale integration. This model mainly used a coarse-to-fine segmentation framework,and further improved the accuracy of the segmentation by combining coarse segmentation at the global scale with fine segmentation at multiple local scales. Experimental results showed that it achieved 82.96% in Dice metrics,and was better than other conventional deep learning methods.It also provides new ideas for segmentation of other tubular organs.
备注/Memo
- 备注/Memo:
-
(收稿日期:2022-01-11)广东省重点领域研发计划项目(2021B0101220006);广东省科技计划项目(2019A050510041);广东省自然科学基金资助项目(2021A1515012300)。△通信作者Email:437871335@qq.com
更新日期/Last Update:
2022-11-08