人工智能在冠状动脉CTA诊断冠状动脉疾病中的应用价值

Application value of artificial intelligence in the diagnosis of coronary artery disease by coronary CTA

目的:探讨人工智能(AI)在冠状动脉(冠脉)CTA(CCTA)诊断冠状动脉疾病中的应用价值。方法:回顾性分析行CCTA和冠脉造影术(CAG)66例(三大主支共198支血管)冠状动脉粥样硬化性心脏病(CHD)患者的影像学资料。CCTA图像分别由高年资主治医师和AI辅助诊断软件进行处理,评价AI辅助诊断软件在冠脉疾病诊断中的可行性及准确性。结果:①AI辅助软件自动对CCTA图像一键式重建分析并生成报告,用时约为人工用时的2.1%。②以CAG为金标准,AI检测冠脉狭窄的敏感度、特异度分别为94.8%、63.6%,检测狭窄程度≥25%的敏感度、特异度分别为90.8%、71.7%;医师诊断冠脉狭窄的敏感度、特异度分别为99.4%、65.9%,检测狭窄程度≥25%的敏感度、特异度分别为96.7%、84.8%;2种诊断方法对冠脉狭窄及狭窄程度≥25%的检出率差异均无统计学意义(均P>0.05)。③以医师诊断为标准,198支血管168个斑块中,AI辅助诊断的敏感度和特异度分别为93.5%和90.0%,AI辅助诊断与医师诊断一致性较高(K=0.752)。④AI辅助诊断还可对冠脉优势型及冠脉支架进行评估,与医师诊断差异均无统计学意义(均P>0.05)。结论:AI辅助诊断在冠脉疾病诊断方面具有较高敏感度和准确性,极大提高了诊断效率,但仍有不足,可作为医师的辅助诊断工具。

Objective:To explore the application value of artificial intelligence(AI)in the diagnosis of coronary artery disease by coronary CTA(CCTA).Methods:The imaging data 198 vessels in three main branches from coronary in 66 patients with coronary heart disease(CHD)who underwent CCTA and coronary angiography(CAG)were analyzed retrospectively.The CCTA images were processed by radiologists and AI-assisted diagnostic software(AI group)to evaluate the feasibility and accuracy of AI-assisted software in the diagnosis of coronary artery disease.Results:①AI-assisted software automatically reconstructed and analyzed CCTA images and generated reports,which took about 2.1%of the manual time.②Taking CAG as gold standard,the sensitivity and specificity of AI in detecting coronary stenosis were 94.8%and 63.6%,in detecting the vessels with stenosis greater than 25%were 90.8%and 71.7%,respectively.The sensitivity and specificity of radiologists in diagnosing coronary stenosis were 99.4%and 65.9%,in diagnosis of the vessels with stenosis greater than 25%were 96.7%and 84.8%,respectively.There were no statistical differences in detection rate of stenosis and vessels with stenosis greater than 25%between AI and radiologists(both P>0.05).③A total of 168 atherosclerotic plaques from 198 blood vessels were diagnosed by radiologists,and taking this results as standard,the sensitivity and specificity of AI detection were 93.5%and 90.0%,respectively,AI-assisted diagnosis had a high diagnostic consistency with radiologists(Kappa=0.752).④AI-assisted diagnosis could assess coronary dominance and coronary stents,which had no statistical differences from radiologists(both P>0.05).Conclusion:AI-assisted diagnosis has a high sensitivity and accuracy in diagnosis of coronary artery disease and improves the diagnostic efficiency,but it is still insufficient and can be used as an auxiliary diagnostic tool.