Computer-aided CT coronary artery stenosis detection: comparison with human reading and quantitative coronary angiography.

Abstract

To evaluate computer-aided stenosis detection for computed tomography coronary angiography (CTA) in comparison with human reading and conventional coronary angiography (CCA) as the reference standard. 50 patients underwent CTA and CCA and out of these 44 were evaluable for computer-aided stenosis detection. The diagnostic performance of the software and of human reading were compared and quantitative coronary angiography (QCA) served as the reference standard for the detection of significant stenosis (>50 %). Overall, three readers with high (reader 1), intermediate (reader 2) and low (reader 3) experience in cardiac CT imaging performed the manual CTA evaluation on a commercially available workstation, whereas the automated software processed the datasets without any human interaction. The prevalence of coronary artery disease was 41 % (18/44) and QCA indicated significant stenosis (>50 %) in 33 coronary vessels. The automated software accurately diagnosed 18 individuals with significant coronary artery disease (CAD), and correctly ruled out CAD in 10 patients. In summary the sensitivity of computer-aided detection was 100 %/94 % (per-patient/per-vessel) and the specificity was 38 %/70 %, the positive predictive value (PPV) was 53 %/42 % and the negative predictive value (NPV) was 100 %/98 %. In comparison, reader 1-3 showed per-patient sensitivities of 100/94/89 %, specificities of 73/69/50 %, PPVs of 72/68/55 % and NPVs of 100/95/87 %. Computer-aided detection yields a high NPV that is comparable to more experienced human readers. However, PPV is rather low and in the range of an unexperienced reader.