Performance standards and edge detection with computerized quantitative coronary arteriography

Quantitative coronary angiography (QCA) has become an important tool for evaluating coronary angiograms. Many methodologic factors, such as the choice of frame to analyze, the selection of the ''normal,'' segment and the method of edge detection used may affect the results of QCA. The sequential steps in performing QCA, including a comparison of visual and automated edge-detection methodologies, were evaluated using 12 precision-drilled phantoms and 20 patient films. Normal diameter, minimal lumen diameter, and diameter stenosis were measured. In the phantom studies, the measurements from both visual and automated systems correlated well with the true measurements of the phantoms and between systems (all r values >0.92). To study the difference between methodologies on QCA results as influenced by the choice of frame and normal segment analyzed, the patient films were analyzed independently in 3 separate rounds of interpretation. In round 1, each system's operator individually chose frames and normal segments for analysis. In round 2, both systems analyzed the same preselected frames, but independently chose normal segments. In round 3, both systems analyzed the same preselected normal segments and frames. The intersystem correlations between visual and automatic systems for rounds 1, 2, and 3 were: normal diameter, r = 0.25, r = 0.37, and r = 0.75, respectively; minimal lumen diameter, r = 0.79, r = 0.86, and r = 0.85, respectively; and diameter stenosis, r = 0.65, r = 0.73, and r = 0.87, respectively. The manual edge-detection and automated edge-detection systems used in this study are reasonably accurate and consistent on phantom studies. In patient studies, the nonautomated processes (choice of frame and normal segment for analysis) produced significant differences in the QCA results, thus illustrating that operator-dependent factors other than edge detection are very important in QCA.