| Objectives:Coronary angiography results was the gold standard.To investigate the accuracy and application of automatic identification of coronary artery stenosis by artificial intelligence.1.To compare the difference between AI and cardiovascular imaging physician in detecting coronary artery segmental stenosis,and the difference in the time required for post-processing CCTA images.2.To evaluate the accuracy of AI in diagnosing different degree of stenosis in CCTA images.Methods:A total of 200 patients with CCTA and CAG from September 2018 to October 2019 in the First Affiliated Hospital of Kunming Medical University were retrospectively collected.Each patient’s coronary artery was divided into 18 segments according to SCCT guidelines.The CCTA images of all patients were postprocessed and diagnosed in AI Group and Doctor Group respectively,and the time required for post-processing and detection of coronary artery stenosis were recorded.According to CAD-RADS classification standard and lumen diameter stenosis≥50%was hemodynamic significance,lumen stenosis was classified into no Stenosis,mild stenosis(<50%),moderate Stenosis(50%～69%),severe stenosis(70%～99%)and occlusion.The coronary artery diameter ≥1.5 mm was included in the diagnostic range.Plaques located outside the lumen that did not cause significant stenosis and occluded segments far from the developing segments were excluded from the evaluation.The diagnosis of coronary angiography was taken as the gold standard,the accuracy of detecting coronary artery stenosis in AI Group and physician group was calculated.The sensitivity,specificity,positive predictive value,negative predictive value and accuracy of the AI group in diagnosing different degree of stenosis were calculated.All the above diagnostic results were compared with coronary angiography for Kappa value consistency test.Results:1.The average post-processing time CCTA in physician group was 322.55±30.60 s,while the average processing time of AI group was 8.05±1.32 s,the difference was statistically significant(Z=-6.28,P<0.05).The average postprocessing time of AI group was 314.50±29.28 s less than that of physician group.2.The sensitivity,specificity,positive predictive value,negative predictive value and accuracy of coronary artery stenosis in AI group were 73.68%-100%87.76%-100%,70%-100%,91.49%-100%and 88.89%-98.96%respectively.3.The sensitivity,specificity,positive predictive value,negative predictive value and accuracy of coronary artery stenosis segments detected in physician group were 30.77%-98.11%,96.81%-100%,90.20%-99.38%,93.43%-99.36%respectively.The stenosis of the dLAD segment detected by the physician was not consistent with CAG(Kappa value 0.45),the stenosis of the remaining segments was found to be consistent with that of CAG.4.The stenosis of dLAD and dRCA in AI group was significantly different from that in physician group(P<0.05),AI group was superior to the physician group.There was no significant difference in 14 segments of coronary artery between the two groups.5.In patients,coronary arteries and segments(lumen stenosis ≥50%),the diagnostic sensitivity,specificity,positive predictive value,negative predictive value and accuracy of AI were 94.76%/90.77%/87.66%,55.56%/96.68%/99.18%,97.84%/96.09%/95.91%,33.33%/92.08%/97.33%,93.00%/93.88%/97.09%respectively.The Kappa values were 0.38,0.88 and 0.89,respectively.6.The overall accuracy of AI group in the diagnosis of coronary artery stenosis was 66.63%,which was similar to that of coronary angiography(Kappa value 0.53).In the diagnosis of mild stenosis and severe stenosis,the consistency between AI group and CAG group was better(Kappa values were 0.62 and 0.68 respectively).In the diagnosis of moderate stenosis,AI group was generally consistent with CAG(Kappa value 0.51).The consistency of AI and CAG was poor which in the diagnosis of occluded segments(Kappa value 0.17).Conclusions:1.The time of the CCTA image processed by AI is shorter than that of the physician,which can improve the work efficiency.2.AI has high accuracy in detecting coronary artery stenosis,and has good or good consistency with CAG.It can assist the physician in the diagnosis and reduce the rate of missed diagnosis,especially for the stenosis of dLAD and dRCA.So AI can be used as a physician-assisted diagnosis.3.The accuracy of AI in the diagnosis of coronary artery stenosis(≥50%)was high,but the accuracy of AI in diagnosing coronary artery stenosis of different degrees was not good,so it still needs to be further improved,especially in the diagnosis of moderate stenosis and occlusion. |
PDF Full Text Request