生命科学仪器2023，Vol.21Issue(6)：82-88,7.DOI:10.11967/2023211218
基于EfficientNet和Grad-CAM++的冠状动脉造影体位识别方法的研究
The Classification of the Positions of Coronary Angiography by EfficientNet and Grad-CAM++
李劼慧 1吴悔 2曾研 3吴水才2
作者信息
- 1. 北京工业大学环境与生命学部,医学电子实验室,北京 100124||国家心血管病中心,国家心血管病重点实验室,中国医学科学院阜外医院 心外科,北京 100037
- 2. 北京工业大学环境与生命学部,医学电子实验室,北京 100124
- 3. 上海联影智能医疗科技有限公司,上海 201807
- 折叠
摘要
Abstract
Objective:Few studies have used artificial intelligence to analyze coronary angiography(CAG),and we propose to achieve automatic recognition of CAG by EfficientNet and Grad-CAM++.Methods:The 86,639 ima-ges were divided into training set,validation set and test set,including all 8 commonly used body positions."Accu-racy"is used as the main index to evaluate the classification effect of the EfficientNet-B1 on each projection posi-tion and Grad-CAM++ is used to explain the learning process of the model.Results:The average accuracy of Ef-ficientNet-B1 on eight imaging positions in the validation set and test set is 90.73%and 89.40%,respectively,which would be higher than VGG-16 and ResNet-18.And Grad-CAM++ shows that EfficientNet-B1 could learn the coronary arteries most comprehensively.conclusion:EfficientNet-B1 can effectively classify the common CAG positions,and the learning effect is better than other network models.关键词
EfficientNet/Grad-CAM++/冠状动脉造影/体位识别Key words
EfficientNet/Grad-CAM++/Coronary Angiography/Classification Model分类
医药卫生引用本文复制引用
李劼慧,吴悔,曾研,吴水才..基于EfficientNet和Grad-CAM++的冠状动脉造影体位识别方法的研究[J].生命科学仪器,2023,21(6):82-88,7.
