计算机科学与探索 ›› 2020, Vol. 14 ›› Issue (9): 1602-1611.DOI: 10.3778/j.issn.1673-9418.1909028

• 图形图像 • 上一篇    下一篇

嵌入注意力机制的多模型融合冠脉CTA分割算法

沈烨,方志军,高永彬   

  1. 上海工程技术大学 电子电气工程学院,上海 201600
  • 出版日期:2020-09-01 发布日期:2020-09-07

Multi-model Fusion of Coronary CTA Segmentation Based on Attention Mechanism

SHEN Ye, FANG Zhijun, GAO Yongbin   

  1. School of Electronic and Electrical Engineering, Shanghai University of Engineering Science, Shanghai 201600, China
  • Online:2020-09-01 Published:2020-09-07

摘要:

冠心病是当今世界上最大的健康问题之一,因此对冠心病的早期预防和诊断非常重要。斑块的存在和冠状动脉血管狭窄是引发冠心病的主要原因,对斑块的检测和冠状动脉血管分割成为检测冠状动脉疾病的首选方案。目前手动分割冠状动脉耗时并且由操作者的主观意识决定,这使得现在的临床医学诊断中对自动分割技术的需要显而易见。提出了一种基于深度学习多模型融合的冠脉CT血管造影(CTA)的血管分割方法,该方法包括三个网络模型:一个原始三维全卷积网络(3D FCN),以及两个在原始3D FCN中嵌入注意力门控(AG)模型的网络。然后把三种网络的预测结果采用多数投票算法进行融合来得到网络预测的最终结果。在后处理阶段,采用水平集函数对网络融合预测的结果进行进一步迭代优化。在评估所提出方法的过程中把JI和DSC分数来作为性能度量进行比较,最终结果表明,所提出的方法提供了更好更准确的分割结果。

关键词: 冠脉CTA, 多模型融合, 三维全卷积网络(3D FCN), 注意力门控模型, 水平集函数

Abstract:

Coronary heart disease is one of the biggest health problems in the world, so the early prevention and diagnosis of coronary heart disease is very important. The presence of plaque and coronary artery stenosis are the main causes of coronary heart disease. The detection of plaque and coronary artery segmentation have become the first choice for detecting coronary artery disease. At present, manual coronary artery segmentation is time-consuming and determined by the operator??s subjective consciousness, which makes the need for automatic segmentation technology obvious in clinical diagnosis. In this paper, a method of computed tomography angiography (CTA) based on deep learning multi-model fusion is proposed, which includes three network models: an original 3D fully convolutional network (3D FCN) and two networks embedded the attention gate (AG) model in the original 3D FCN. And then the prediction results of the three networks are fused by majority voting algorithm to obtain the final result of the network prediction. In the post-processing stage, the level set function is used to further iteratively optimize the network fusion prediction results. In the process of evaluating the proposed method, Jaccard index (JI) and Dice similarity coefficient (DSC) scores are compared as performance measures. The final results show that the proposed method provides better and more accurate segmentation results.

Key words: coronary computed tomography angiography (CTA), multi-model fusion, three-dimensional fully convolutional network (3D FCN), attention gate model, level set function