摘要
冠心病作为高发病率的重大冠状动脉疾病,让影像科医生承担了繁重的工作强度。利用人工智能识别冠状动脉中的斑块,实现对冠心病血管狭窄程度的判断,可以减轻CT影像医生的工作强度。应用增加了注意力机制的U-Net网络对冠状动脉进行识别分割,可以大大减少大夫的诊断时间。应用增加了注意力机制的U-Net网络进行训练,最终使损失函数、准确率Accuracy以及Dice指标分别达到0.021 3、0.962 6、0.999 9。同时通过预测的结果可以看出效果令人满意。对后续斑块的定性分析、钙化积分计算等提供了良好的分割结果。
As a major coronary artery disease with high incidence rate, coronary heart disease has increased the working intensity of radiology doctor. Identifying plaque in coronary artery and judging degree of coronary artery stenosis by using artificial intelligence can reduce work intensity of CT imaging doctors. The U-Net network with attention mechanism is used to recognize and segment coronary artery which can greatly reduce doctor’s diagnosis time. The application of U-Net with attention mechanism in model training can minimize loss function to 0.021 3 and increase the accuracy and the dice to 0.962 6 and 0.999 9 respectively. This method provides good segmentation results for qualitative analysis and calcification score calculation of following plaques.
作者
冯雪聪
陈波
钱俊磊
曾凯
陈伟彬
李晓琳
潘红红
FENG Xuecong;CHEN Bo;QIAN Junlei;ZENG Kai;CHEN Weibin;LI Xiaolin;PAN Honghong(College of Electrical Engineering,North China University of Science and Technology,Tangshan HeBei 063210,China)
出处
《激光杂志》
CAS
北大核心
2022年第2期200-204,共5页
Laser Journal
基金
河北省省属高等学校基本科研业务费研究项目(No.JYG2020004):基于深度学习的结肠癌辅助诊断的研究
华北理工大学教育教学改革研究与实践项目(No.L20121)。
关键词
深度学习
冠心病
U-Net
注意力机制
deep learning
coronary artery disease
U-Net
attention mechanism
