Avan Suinesiaputra1, Patrick J H de Koning1, Elena Zudilova-Seinstra2, Johan H C Reiber1, Rob J. van der Geest1
1Div. of Image Processing, Dept. of Radiology, Leiden University Medical Center, Leiden, Netherlands; 2Section of Computational Science, University of Amsterdam, Amsterdam, Netherlands
A 3D segmentation method for MRA images is presented. The algorithm uses Non-Uniform Rational B-Splines model of tubular structure. The models surface is fitted based on image forces (searching the closest edge) and model forces (keeping the tubular model). Ten patients with indication of carotid arterial diseases were selected. The method was compared on axially sliced 2D expert-drawn contours (831 slices). The average overlapped contour area was 65.18% (SD 10.66%). The correlation coefficient for the cross-sectional area was 0.83 (p < 1e-16). The mean cross-sectional area difference was 2.310.1 mm2.