人工智能在冠状动脉CT血管成像对冠心病诊断效能的研究  

Study on the diagnostic efficacy of artificial intelligence for coronary CT angiography in coronary heart disease

作　　者：赵艳红[1] 苏治祥 张晓文 曹永佩 哈若水[1] ZHAO Yanhong;SU Zhixiang;ZHANG Xiaowen;CAO Yongpei;HA Ruoshui(Medical Imaging Center,the People's Hospital of Ningria Hui Autonomous Region,Yinchuan 750002,China)

机构地区：[1]宁夏回族自治区人民医院医学影像中心,宁夏银川750002

出　　处：《实用放射学杂志》2023年第4期557-560,共4页Journal of Practical Radiology

基　　金：宁夏自然科学基金项目(2021AAC03316);宁夏回族自治区重点研发项目(2021ZDYF0148)。

摘　　要：目的探讨人工智能(AI)在冠状动脉CT血管成像(CTA)对冠状动脉狭窄及斑块性质的诊断价值。方法收集临床疑似冠心病行冠状动脉CTA检查及在2周内行选择性冠状动脉造影术(CAG)的患者50例。利用AI冠状动脉图像CT血管狭窄分析软件进行冠状动脉图像诊断,以CAG检查结果为金标准,评价AI对冠状动脉狭窄诊断的敏感度、特异度、阳性预测值、阴性预测值及准确度,评价AI对冠状动脉不同狭窄程度及不同斑块性质诊断的敏感度。结果50例患者可评价血管节段为769个,AI对冠状动脉狭窄总体检出的敏感度、特异度、阳性预测值、阴性预测值及准确度分别为89.80%(273/304)、98.92%(460/465)、98.20%(273/278)、93.69%(460/491)及95.32%(733/769)。AI对冠状动脉轻度狭窄、中度狭窄、重度狭窄及闭塞诊断的敏感度分别为84.97%(130/153)、85.71%(78/91)、81.13%(43/53)、71.43%(5/7),AI对冠状动脉闭塞性病变检出的敏感度低于轻度、中度及重度狭窄(P<0.05)。AI对钙化斑块、非钙化斑块及混合斑块检出的敏感度分别为92.11%(140/152)、73.87%(82/111)、91.78%(67/73),对钙化斑块及混合斑块的检出的敏感度优于非钙化斑块(P<0.05)。结论AI在冠状动脉CTA中对冠心病的诊断具有较高的敏感度及特异度。AI对冠状动脉闭塞性病变检出的敏感度低于轻度、中度及重度狭窄,对钙化斑块及混合斑块检出的敏感度优于非钙化斑块。Objective To explore the value of artificial intelligence(Al)in the diagnosis of coronary stenosis and nature of the plaque by coronary CT angiography(CTA).Methods A total of 50 patients with clinically suspected coronary heart disease underwent coronary CTA and underwent selective coronary arteriography(CAG)within two weeks were collected.AI coronary image CT vascular stenosis analysis software was used to coronary image diagnosis.The results of CAG were used as the gold standard to evaluate the sensitivity,specificity,positive predictive value,negative predictive value and accuracy of Al in the diagnosis of coronary stenosis,and to evaluate the sensitivity of Al in the diagnosis of different degree of coronary stenosis and different nature of plaque.Results A total of 769 vascular segments were evaluated in 50 patients.The sensitivity,specificity,positive predictive value,negative predictive value and accuracy of AI in the overall detection of coronary stenosis were 89.80%(273/304),98.92%(460/465),98.20%(273/278),93.69%(460/491)and 95.32%(733/769),respectively.The sensitivity of AI to the diagnosis of mild,moderate,severe coronary stenosis and occlusion were 84.97%(130/153),85.71%(78/91),81.13%(43/53),and 71.43%(5/7),respectively.The sensitivity of AI to coronary occlusion was lower than that to mild,moderate and severe stenosis(P<0.05).The sensitivity of AI to calcified plaques,non-calcified plaques and mixed plaques were 92.11%(140/152),73.87%(82/111)and 91.78%(67/73),respectively.The sensitivity of AI to calcified plaques and mixed plaques was better than that to non-calcified plaques(P<o.05).Conclusion AI diagnosis using in coronary CTA has high sensitivity and specificity in coronary heart disease.The sensitivity of AI in detecting coronary occlusive lesions is lower than that in detecting mild,moderate and severe stenosis,and the sensitivity of AI in detecting calcified plaques and mixed plaques is better than that in detecting non-calcified plaques.

关 键 词：人工智能 冠状动脉 计算机体层成像 血管造影术 冠心病 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R814.43[自动化与计算机技术—控制科学与工程] R543.3[医药卫生—影像医学与核医学]