Comparison of Hemodynamic Significant Coronary Stenosis Using Artificial Intelligence and Conventional Cross Sectional Area in CT Coronary Angiography

Abstract

<i>Objectives</i>: This was study was designed to compare the accuracy of estimation of coronary stenosis in patients with stable coronary artery disease using artificial intelligence based technique and conventional cross sectional area method using hemodynamically significant stenosis based on CTFFR as the gold standard for significant ischemia. <i>Background</i>: Although detection of degree of stenosis on coronary angiography as well Computed tomographic angiography (CTA) forms the backbone of management plan of a patient with suspected coronary artery disease there is a discordance in the results between both the techniques for estimation of stenosis as well as 20-30% interobservor variation in the results in the stenosis estimation based on conventional angiography method. So a more robust method is needed using modern techniques like artificial intelligence to address this problem. <i>Methods</i>: CTA’s of 100 consecutive patients of stable coronary artery disease were evaluated for coronary stenosis on per vessel and per patient basis using conventional cross section method and using artificial intelligence with hemodynamic significant stenosis using CT FFR < 0.8 as the gold standard for ischemia and the results compared. <i>Results</i>: Cross sectional area method revealed significant stenosis > 50% in 184 (61.2%) vessels per vessel basis and in 65% on per patient basis while the AI method showed significant stenosis in all the 232 (77.2%) per vessels and in 89% per patient basis out of which 81% were hemodynamically significant. 18% of cases were not assessable by cross section method due to heavy vessel calcifications. Sensitivity and specificity on per patient basis by cross section method and AI method were 80%, 57% and 98% and 90% respectively with a false negative and positive of 19%, 42% by cross section method and 1.3%, 10% by AI method respectively with overall accuracy of 76% and 97% respectively of the two methods. <i>Conclusion</i>: AI method of estimation of coronary stenosis is more accurate and robust than conventional area estimation method in clinical practice especially when patients have higher vessel coronary calcium.