| Objective To explore the diagnostic efficacy of Artificial Intelligence,AI)in interpreting coronary computed tomography angiography(CCTA)for coronary atherosclerotic stenosis.Methods(1)We retrospectively collected the data of patients who underwent CCTA examination and Coronary Angiography(CAG)examination within 2 weeks in our hospital from January 2018 to September 2019.In strict accordance with the inclusion criteria and exclusion criteria,a total of 90 patients(58 males,32 females,)were included in this study.(2)All CCTA original thin-layer images included in the study cases were pushed to GE image post-processing workstation,and two doctors with 8 years of coronary artery post-processing and diagnosis experience performed image processing,result diagnosis of artificial coronary artery analysis software,and automatic coronary artery image post-processing and result diagnosis by Coronary Doc 1.0,The observation indexes include whether AI correctly identifies coronary artery branches,the total time of post-processing and diagnosis of two ways,the degree of stenosis and the nature of plaque.The diagnosis results are respectively taken as "artificial group" and "AI group".Two cardiologists who have been engaged in coronary angiography for more than 5 years respectively completed CAG results analysis(including stenosis site and degree)by single blind method,and the diagnosis results were taken as "CAG group".(3)we compare the timeline of AI and manual post-processing CCTA and diagnosis.In addition,referring to manual interpretation,the accuracy of AI in identifying coronary artery branches was calculated.(4)Firstly,the sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate for identifying coronary atherosclerotic stenosis and diagnosing coronary heart disease in AI group and artificial group were calculated respectively based on the number of coronary segments with CAG group as the "gold standard".Secondly,the consistency of AI group and CAG group,artificial group and CAG group in identifying coronary atherosclerotic stenosis and diagnosing coronary heart disease is compared respectively,so as to evaluate the diagnostic efficacy of AI and artificial methods in diagnosing coronary atherosclerotic stenosis and coronary heart disease respectively.(5)The sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate,and Kappa value of AI group in identifying coronary atherosclerotic stenosis and diagnosing coronary heart disease were calculated with artificial group as reference,and the diagnostic efficiency and consistency of AI in diagnosing coronary atherosclerotic stenosis and coronary heart disease were evaluated compared with manual interpretation of CCTA.(6)With artificial group as reference,based on the number of plaques with three different properties(calcified plaque,mixed plaque and non-calcified plaque),the overall accuracy rate of AI identification plaque and the respective accuracy rates of the three types of plaques are calculated,and the consistency of the identification plaque properties between AI group and artificial group is compared,so as to evaluate the diagnostic efficiency of AI on plaque properties.Results(1)The total time of CCTA post-treatment and diagnosis in AI group and artificial group were 317.94±60.44 s,481.40±66.35 s respectively,t=21.815,with ?=0.05 as the test level,the difference was statistically significant(P=0.000).(2)In 90 patients,there were 524 coronary artery branches,and the total number of coronary artery branches with wrong AI identification marks was 13,with an accuracy rate of 97.52%(511/524).(3)Among the 90 patients enrolled in the study,the artificial group,AI group and CAG group analyzed 1062 segments of coronary artery segments respectively.Taking CAG group as "gold standard",artificial group correctly diagnosed 357 segments of 381 segments of coronary artery stenosis and 213 segments of 235 segments of significant coronary artery stenosis(stenosis ?50%).The sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate of artificial interpretation CCTA in diagnosing coronary atherosclerotic stenosis were 93.70%(357/381),94.13%(641/681),96.39%(641/665),89.92%(357/397)and93.97%(998/1062),respectively.The sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate of manual interpretation CCTA in diagnosing coronary heart disease are 90.64%(213/235),97.10%(803/827),97.33%(803/825),89.87%(213/237)and 95.67%(1016/1062),respectively.The diagnosis of coronary atherosclerotic stenosis and coronary heart disease in artificial group and CAG group have excellent consistency(Kappa values are 0.870 and 0.875 respectively),and the difference is not statistically significant(P values are 0.060 and 0.883 respectively).(4)Taking CAG group as the "gold standard",AI group correctly diagnosed 319 segments of 381 segments of coronary artery stenosis and 128 segments of 235 segments of coronary artery stenosis.The sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate of AI interpretation CCTA in diagnosing coronary atherosclerotic stenosis are83.73%(319/381),90.90%(619/681),90.90%(619/681),83.73%(319/381),88.32%(938/1062),respectively.The sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate of AI in diagnosing coronary heart disease were 54.47%(128/235),98.31%(813/827),88.37%(813/920),90.14%(128/142)and88.61%(941/1062),respectively.The consistency of diagnosis of coronary atherosclerotic stenosis between AI group and CAG group is good(Kappa=0.746),and the difference is not statistically significant(P= 1.000).The consistency of diagnosis of coronary heart disease between AI group and CAG group is also good(Kappa=.615),and the difference between the two groups is statistically significant(P=0.000).(5)Taking the artificial group as a reference,AI group correctly diagnosed 340 of397 coronary stenosis segments,and AI interpreted CCTA's sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate for diagnosing coronary atherosclerotic stenosis were 85.64%(340/397),93.83%(624/665),91.63%(624/681),89.24%(340/381)and 90.77%(964/1062),respectively.The consistency between AI group and artificial group in diagnosing coronary atherosclerotic stenosis is excellent(Kappa=0.801),and the difference is not statistically significant(P=0.129).Secondly,the sensitivity,specificity,negative predictive value,positive predictive value and coincidence rate of AI group in diagnosing coronary heart disease are 58.65%(139/237),99.64%(822/825),89.35%(822/920),97.89(139/142)? 90.49%(961/1062),respectively.The consistency of diagnosis of coronary heart disease between AI group and CAG group is good(Kappa=0.680),and the difference between the two groups is statistically significant(P=0.000).(6)493plaques were detected manually,486 plaques were detected in AI group,of which 418 plaques of different nature were accurately identified in AI group,with an overall accuracy rate of84.79%(418/493),174 calcified plaques were correctly identified in AI group,with an accuracy rate of 92.06%(174/189),110 mixed plaques were correctly identified in AI group,with an accuracy rate of 77.47%(110/142).The AI group correctly identified 134non-calcified plaques with an accuracy of 82.72%(134/162),while the AI group missed 41 plaques with a missed diagnosis rate of 8.32%(41/493)and 34 misdiagnosed plaques with a misdiagnosis rate of 6.90%(34/493).AI and artificial interpretation methods have good consistency in identifying plaque properties(K=0.705).Conclusion(1)The Coronary Doc based on deep learning is superior to manual group in CCTA image post-processing and diagnosis speed,and the accuracy of identifying coronary artery branches is extremely high,which can reduce the workload of doctors in coronary artery post-processing.(2)Based on the results of CAG group,although the currently trained AI has insufficient efficiency in diagnosing coronary heart disease,it has higher efficiency in diagnosing coronary atherosclerotic stenosis,and the identification of plaque properties has better consistency and higher accuracy with the artificial group.It has certain clinical value in diagnosing coronary atherosclerotic stenosis,assisting radiologists in processing and interpreting CCTA,reducing the working pressure of radiologists,and improving the working mode. |
PDF Full Text Request