Advanced contour detection for three-dimensional intracoronary ultrasound: a validation – in vitro and in vivo

Publisher Website
Google Scholar
Abstract
Intracoronary ultrasound (ICUS) provides high-resolution transmural images of the arterial wall. By performing a pullback of the ICUS transducer and three-dimensional reconstruction of the images, an advanced assessment of the lumen and vessel wall morphology can be obtained. To reduce the analysis time and the subjectivity of boundary tracing, automated segmentation of the image sequence must be performed. The Quantitative Coronary Ultrasound – Clinical Measurement Solutions (QCU-CMS) (semi)automated analytical software package uses a combination of transversal and longitudinal model and knowledge-guided contour detection techniques. On multiple longitudinal sections through the pullback stack, the external vessel contours are detected simultaneously, allowing mutual guidance of the detection in difficult areas. Subsequently, luminal contours are detected on these longitudinal sections. Vessel and luminal contour points are transformed to the individual cross-sections, where they guide the vessel and lumen contour detection on these transversal images. The performance of the software was validated stepwise. A set of phantoms was used to determine the systematic and random errors of the contour detection of external vessel and lumen boundaries. Subsequently, the results of the contour detection as obtained in in vivo image sets were compared with expert manual tracing, and finally the contour detection in in vivo image sequences was compared with results obtained from another previously validated ICUS quantification system. The phantom lumen diameters were underestimated by 0.1 mm, equally by the QCU-CMS software and by manual tracing. Comparison of automatically detected contours and expert manual contours, showed that lumen contours correspond very well (systematic and random radius difference: −0.025 ± 0.067 mm), while automatically detected vessel contours slightly overestimated the expert manual contours (radius difference: 0.061 ± 0.037 mm). The cross-sectional vessel and lumen areas as detected with our system and with the second computerized system showed a high correlation (r = 0.995 and 0.978, respectively). Thus, use of the new QCU-CMS analytical software is feasible and the validation data suggest its application for the analysis of clinical research.