Automated morphometry of coronary arteries with digital image analysis of intravascular ultrasound

We designed and tested digital image processing strategies to perform fully automated segmentation of luminal and medial-adventitial boundaries in intravascular ultrasound images of human coronary arteries. Automated segmentation is an essential tool for advanced techniques of clinical visualization and quantitative measurement. Vascular compliance measurements and three-dimensional reconstructions are demonstrated as examples of such applications. Digital image processing was performed in three phases: (1) preprocessing, including a polar transform, local contrast enhancement, and speckle noise filtering; (2) segmentation, involving radial scanning, region growing, or cost-function minimization techniques; and (3) postprocessing, involving dropout filtering and outline smoothing. Cross-sectional areas were compared with manual tracings from experienced operators and showed good agreement. The algorithm bias ranged from -0.34 to 1.18 mm2; interclass and intraclass correlation coefficients ranged from 0.83 to 0.94. The designed techniques currently allow fully automated segmentation without operator interaction of the luminal and, if present, medial-adventitial boundary.