Feasibility of using deep learning to detect coronary atherosclerotic heart disease based on palm images
于文娟 1姚旭婧 2赵文菊 1赵蓓 1姚刚 1刘新艳
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 1南京医科大学第二附属医院中医科，南京 210003
2. 2英国莱斯特大学计算机和数学科学学院，莱斯特 LE17RH
折叠
摘要
目的 探讨基于手掌图像的深度学习检测冠状动脉粥样硬化性心脏病（冠心病）的可行性。 方法 选取2021年9月至2022年5月在南京医科大学第二附属医院心脏内科行择期冠状动脉造影检查的患者，其中冠状动脉狭窄>75%组54例，冠状动脉无狭窄（对照）组38例，共获得184张手掌图像，冠状动脉狭窄>75%组手掌图像108张，冠状动脉无狭窄组手掌图像76张。数据集分为2类，在每个类别中，随机选取80%作为训练集，另外20%作为测试集。采用新型集成学习模型AdaD-IRV2对冠心病数据集进行研究分析，使其能够对输入的人类手掌图像的类别作出快速、自动、相对准确的诊断。 结果 基于手掌图像的冠心病检测算法AdaD-IRV2的平均灵敏度为84.89%，平均准确度为72.82%，平均精确度为72.96%，特异度为50.5%。受试者工作特征曲线下面积（AUC）为0.825。 结论 基于手掌图像的深度学习算法有助于冠心病的检测，该技术有望用于门诊或社区的冠心病筛查。 Objective To explore the feasibility of deep learning based on palm images to detect coronary atherosclerotic heart disease (CHD) . Methods Patients undergoing elective coronary angiography in the Department of Cardiology in the Second Affiliated Hospital of Nanjing Medical University from September 2021 to May 2022 were selected, including 38 patients in the coronary stenosis free (control) group and 54 patients in the coronary stenosis >75% group. A total of 184 palm images were obtained, 108 images in the group with coronary stenosis>75%, and 76 images in coronary stenosis free. The data set was divided into 2 categories, and for each category, 80% were randomly selected as the training set, and the other 20% served as the testing set. We used the new ensemble learning model AdaD-IRV2 to research and analyse the coronary heart disease data set, enabling it to quickly, automatically, and relatively accurately diagnose the categories of input human palm images. Results Algorithmic test results showed that the palm image-based CHD detection algorithm AdaD-IRV2 had an average sensitivity of 84.89%, average accuracy of 72.82% and average precision rate of 72.96%. The specificity was 50.5%. The area under the subject operating characteristic curve (area under curve, AUC) was 0.825. Conclusion Deep learning algorithms based on palm images facilitate the detection of CHD is expected to be used for CHD screening in outpatient or community.
Abstract
Objective To explore the feasibility of deep learning based on palm images to detect coronary atherosclerotic heart disease (CHD) . Methods Patients undergoing elective coronary angiography in the Department of Cardiology in the Second Affiliated Hospital of Nanjing Medical University from September 2021 to May 2022 were selected, including 38 patients in the coronary stenosis free (control) group and 54 patients in the coronary stenosis >75% group. A total of 184 palm images were obtained, 108 images in the group with coronary stenosis>75%, and 76 images in coronary stenosis free. The data set was divided into 2 categories, and for each category, 80% were randomly selected as the training set, and the other 20% served as the testing set. We used the new ensemble learning model AdaD-IRV2 to research and analyse the coronary heart disease data set, enabling it to quickly, automatically, and relatively accurately diagnose the categories of input human palm images. Results Algorithmic test results showed that the palm image-based CHD detection algorithm AdaD-IRV2 had an average sensitivity of 84.89%, average accuracy of 72.82% and average precision rate of 72.96%. The specificity was 50.5%. The area under the subject operating characteristic curve (area under curve, AUC) was 0.825. Conclusion Deep learning algorithms based on palm images facilitate the detection of CHD is expected to be used for CHD screening in outpatient or community.
万方数据