Abstract 211: Automated Computer Aided Stenosis Detection at Coronary CT Angiography -Initial Experience
PURPOSE: To evaluate the performance of a computer aided algorithm for automated stenosis detection at coronary CT angiography (cCTA) using quantitative coronary angiography (QCA) as the reference standard.
METHODS: 59 patients (38 men, mean age 58±12y) underwent cCTA and QCA. In 29 patients cCTA was performed using dual-source CT (Definition™, Siemens, Forchheim, Germany) and in 30 patients using 64-slice CT (Sensation 64™, Siemens). All cCTA data sets were analyzed using a software algorithm (COR Analyzer™, Rcadia, Haifa, Israel) aimed at automated detection of coronary artery stenosis. The performance of the automated algorithm for detection of significant (≥50%) stenosis was compared with QCA based on a 10-segment coronary model. Performance of the automated algorithm was determined on a per-vessel and per-patient basis.
RESULTS: All cCTA data sets were successfully processed by the automated algorithm. QCA revealed a total of 40 stenoses ≥50% of which the software application correctly identified 28 (70%). Overall, the automated algorithm had 70%/90% sensitivity, 83%/63% specificity, 46%/58% PPV, and 93%/92% NPV for diagnosing significant stenosis on per-vessel/per-patient analysis, respectively, compared with QCA. There were a total of 33 false positive detection marks (FP). Upon visual evaluation, 18/33 FP were associated with stenotic lesions <50% on QCA.
CONCLUSION: Compared with QCA, the automated algorithm evaluated in this study has relatively high accuracy for diagnosing significant stenosis at cCTA. If used as a second reader, the high NPV may further enhance the confidence of excluding significant stenosis based on a normal or near-normal cCTA study.