Computer-aided stenosis detection at coronary CT angiography: effect on performance of readers with different experience levels

Objectives: To evaluate the effect of a computer-aided detection (CAD) algorithm for coronary CT angiography (cCTA) on the performance of readers with different experience levels. Methods: We studied 50 patients (18 women, 58 ± 11 years) who had undergone cCTA and quantitative coronary angiography (QCA). Eight observers with varying experience levels evaluated all studies for ≥50 % coronary artery stenosis. After 3 months, the same observers re-evaluated all studies, this time guided by a CAD system. Their performance with and without the CAD system (sensitivity, specificity, positive predictive value and negative predictive value) was assessed using the Likelihood Ratio Χ² test both at the per-patient and per-vessel levels. Results: The sensitivity of the CAD system alone for stenosis detection was 71 % per-vessel and 100 % per-patient. There were 54 false positive (FP) findings within 199 analyzed vessels, most of them associated with non-obstructive (<50 %) lesions. With CAD, one (out of three, 33 %) inexperienced reader’s per-patient sensitivity and negative predictive value significantly improved from 79 % to 100 % (P = 0.046) and from 90 % to 100 % (P = 0.034), respectively. Other readers’ performance indices showed no statistically significant change. Conclusions: Our results suggest that CAD can improve some inexperienced readers’ sensitivity for diagnosing coronary artery stenosis at cCTA. Key Points: • Computer-assisted algorithms can aid in detecting coronary artery stenosis at CT. • The change in overall diagnostic performance is small across all experience levels. • For some inexperienced readers a significant increase in sensitivity is observed.