| 李浚利,黄益龙,韩丹,蔡雅倩,闵蕊,刘顼,康绍磊,段慧.冠状动脉CT血管成像中人工智能诊断冠心病的准确性[J].中国医学影像技术,2021,37(1):59~62 |
| 冠状动脉CT血管成像中人工智能诊断冠心病的准确性 |
| Diagnostic accuracy of artificial intelligence for coronary heart disease in coronary CT angiography |
| 投稿时间：2019-09-11 修订日期：2020-09-30 |
| DOI：10.13929/j.issn.1003-3289.2021.01.013 |
| 中文关键词: 冠状动脉疾病 人工智能 冠状动脉造影 体层摄影术,X线计算机 |
| 英文关键词:coronary disease artificial intelligence coronary angiography tomography, X-ray computed |
| 基金项目:云南省基础研究计划昆医联合专项基金（2018FE001-208）。 |
|
| 摘要点击次数: 2116 |
| 全文下载次数: 688 |
| 中文摘要: |
| 目的 观察冠状动脉CT血管成像（CCTA）中人工智能（AI）诊断冠心病（CHD）的准确性。方法 回顾性分析105例临床拟诊CHD患者间隔2周内CCTA及有创冠状动脉造影（ICA）资料。以ICA结果为金标准，分别以患者、冠状动脉及其节段为单位，计算AI判断冠状动脉狭窄程度≥ 50%的敏感度、特异度、阳性预测值、阴性预测值和准确率，并对AI与ICA结果进行一致性检验。结果 ①以患者为单位，AI诊断敏感度、特异度、阳性预测值、阴性预测值和准确率分别为97.92%（94/96）、66.67%（6/9）、96.91%（94/97）、75.00%（6/8）和95.24%（100/105）；②以冠状动脉为单位，AI诊断敏感度、特异度、阳性预测值、阴性预测值和准确率分别为91.05%（173/190）、98.26%（226/230）、97.74%（173/177）、93.00%（226/243）和95.00%（399/420），与ICA一致性好；③以血管节段为单位，AI诊断敏感度、特异度、阳性预测值、阴性预测值和准确率分别为83.90%（224/267）、97.67%（1215/1244）、88.54%（224/253）、96.58%（1215/1258）和95.23%（1439/1511）。各层面AI与ICA的一致性分别为较好、好和较好（Kappa=0.68、0.90和0.83，P均<0.001）。结论 CCTA图像质量较好且AI能正确识别冠状动脉血管树时，其诊断CHD准确率较高，可作为辅助诊断工具。 |
| 英文摘要: |
| Objective To explore the accuracy of artificial intelligence (AI) in diagnosis of coronary heart disease (CHD) in coronary artery CT angiography (CCTA). Methods Data of coronary artery imaging of 105 patients with suspected CHD undergoing CCTA and invasive coronary angiography (ICA) within 2 weeks, and the latter was taken as gold standard. The sensitivity, specificity, positive predictive value, negative predictive value and accuracy of AI for diagnosing coronary artery stenosis ≥ 50% were calculated. The consistency between AI and ICA results was observed. Results ①At the patient level, the sensitivity, specificity, positive predictive value, negative predictive value and accuracy of AI for diagnosing CHD was 97.92% (94/96), 66.67% (6/9), 96.91% (94/97), 75.00% (6/8) and 95.24% (100/105), respectively.②At the level of coronary arteries, the sensitivity, specificity, positive predictive value, negative predictive value and accuracy was 91.05% (173/190), 98.26% (226/230), 97.74% (173/177), 93.00% (226/243) and 95.00% (399/420), respectively. ③At the lesion segment level, the sensitivity, specificity, positive predictive value, negative predictive value and accuracy was 83.90% (224/267), 97.67% (1215/1244), 88.54% (224/253), 96.58% (1215/1258) and 95.23% (1439/1511), respectively. The consistency between AI and ICA was good at patient level and segment level(Kappa=0.68, 0.90), better at coronary artery level (Kappa=0.83, all P<0.001). Conclusion When the quality of CCTA imaging was good and AI could correctly identify coronary artery tree, AI had good diagnostic accuracy, which could be used as an auxiliary tool for CHD. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
