Three-dimensional automated segmentation of coronary wall and plaque from intravascular ultrasound pullback sequences

Abstract

An automated method for segmentation of intravascular ultrasound (IVUS) pullback image sequences is reported. The method uses contextual information about positioning of coronary borders in the previous frames of the sequence to estimate the region of interest in which the internal and external elastic laminae and plaque borders are determined using a graph searching approach. The method was applied to IVUS pullback image sequences acquired in vivo. Systolic and mid-diastalic EGG-gated sequences were derived from each pullback and automatically analyzed in both directions. The method successfully identified the borders in all 20 image sequences. A good correlation was obtained between computer-detected and observer-defined original lumen and plaque areas (r=0.98, y=1.0x+11.51; r=0.94, y=1.06x-0.09; respectively). Border positions were quite accurate with a root-mean-square border positioning error of 0.15/spl plusmn/0.07 mm. The authors' results demonstrate the feasibility of automated border detection in EGG-gated IVUS pullback image sequences.