A deterministic approach to automated stenosis quantification

Abstract

We developed a new approach to quantitative coronary angiography (QCA), which overcomes several limitations of available programs, such as dependence on operator input; limited tracking ability; fixed correction of the point spread function (PSF), and different calibration on empty vs. contrast-filled catheters. The program (Intelligent Images QCA, version 1.4) provides absolute reproducibility by deterministic, operator-independent identification of the skeleton and the edges of the coronary tree. The algorithm works as follows: application of a matched filter to emphasize selectively the coronary arteries; adaptive threshold binarization; binary thinning and skeletonization; perpendicular resampling with sub-pixel interpolation; derivative filtering; minimal cost edge detection; and automatic identification and quantification of the stenosis. Operator's interaction is restricted to definition of a region of interest; editing of either skeleton or edges is not allowed. PSF correction is fine-tuned to the actual frequency response of the imaging chain by calibration on a contrast- filled conical lucite phantom. Catheter calibration is carried out by a second derivative-based edge detection much less sensitive to the presence of contrast. In vitro phantom analysis (0.5 to 5.0 mm) showed accuracy of 0.028- 0.031 mm and precision of 0.054-0.062 mm on nonmagnified images from the angio TV chain and the cine projector, respectively. In vivo evaluation on a series of consecutive diagnostic angiograms yielded correct contour detection of 70/73 stenoses (96%); interobserver intraframe MLD variability 0.00 mm; correct tracking of catheter edges 100%; interobserver variation coefficient of catheter calibration 3.3%; and mean difference of calibration factor on contrast-filled vs. empty catheters 2.7%. This new approach significantly improves reproducibility with respect to conventional QCA, maintaining high accuracy, precision, and applicability.