Appendix of the Manuscript: Automated Quantification
of Coronary Plaque with Computed Tomography:
Comparison with IVUS using a Dedicated Registration
Algorithm for Fusion-Based Quantification
IVUS
Lumen and Vessel Wall Contour Detection
Coronary plaque characteristics were evaluated on intravascular ultrasound (IVUS) data sets by
an independent and blinded observer using offline dedicated post-processing software (IVUS
CMS 4.0, Medis Medical Imaging Systems, Leiden, the Netherlands). Along the full length of the
IVUS run, lumen-intima and media-adventitia interface were identified in longitudinal and
transversal views by an automatic contour detection algorithm. A limited number of individual
cross-sectional frames were manually adapted to optimize lumen and vessel wall contour
detection. IVUS contour detection was performed independently from the contour detection
derived from CT imaging.
Computed Tomography
Lumen and Vessel Wall Contour Detection
On cardiac computed tomography (CT) data sets, the dedicated software was able to detect both
lumen and vessel wall contours which were used for automated quantitative measurements of
coronary plaques, as depicted in Figure 1, upper panel. At first, a fast vessel-tracking algorithm
was used to obtain the 3-dimensional centerline (ranging from the proximal to distal marker) of
each coronary artery. This vessel-tracking step consists of: (1) a pre-segmentation of the vessel
between the proximal and distal point and (2) a fast path backtracking from distal to the proximal
point through the center of the segmentation. Based on this centerline, a stretched multi-planar
reformatted (MPR) volume was created of the segment of interest. MPR volumes allow the
analysis of curved coronary arteries as straight vessels. Next, four longitudinal cross-sections
were extracted from the MPR volume at 45 degrees angular intervals. Subsequently, lumen
borders in these four longitudinal images were detected by a model guided minimum cost
approach (MCA). The MCA method uses a combination of spatial first-, and second-derivative
gradient filters in combination with knowledge of the expected CT intensity values in the arteries.
Therefore, the MCA method is not sensitive to differences in attenuation values between different
data sets. The longitudinal detection allows a smooth interpolation between the proximal and
distal sites of coronary bifurcations. The MCA with a circular lumen model was used to detect the
lumen border contours in each transversal slice of the MPR volume. In this step, the intersection
points of each transversal slice with the earlier obtained longitudinal contours were used to guide
the contour detection.
Additionally, the vessel wall borders were detected in these longitudinal images by a similar
MCA with a different model. The applied model was based on several pre-defined constraints
concerning the location of longitudinal vessel wall contours; contours were positioned outside the
detected longitudinal lumen contours and the regions with high intensity values (e.g. calcified
regions) were included according to a relative weighting scheme. Consecutively, transversal
vessel wall contours were fitted on the transversal slices using the intersection of the longitudinal
contours with each slice as attraction point. These automated processing steps were independent
from the standard viewing settings (window level 1024, width 0).