Evaluation of segmentation algorithms for vessel wall detection in echo particle image velocimetry

Recent in-vitro and in-vivo validation studies confirmed the accuracy of echo particle image velocimetry (echo PIV), a simple non-invasive means of measuring multi-component blood velocity vectors. Echo PIV should also be useful for direct measurement of wall shear stress (WSS) in clinical studies. However, calculation of WSS requires accurate delineation of vessel walls in ultrasound images, which may be problematic when conventional segmentation techniques are used. In this paper, we proposed two methods for segmenting contrast enhanced B-mode images. The first is based on the intensity profile of ultrasound images, termed intensity-based edge detection (IBED) and the second based on the movement of microbubbles, termed movement-based quadratic difference (MBQD). The parameters related with the two methods were optimized over large sets of microbubble images acquired from human carotid vessels using an echo PIV system (Illumasonix LLC, Boulder, CO). A validation study on the two algorithms was carried out against manual delineations on both common carotid artery (CCA) and carotid bifurcation images, with 20 frames for each group. The inter-observer variability of three manual delineations, in pixels (about 80 ¿m/pixel), was 0.9±0.4, 1.3±0.6, 1.3±0.6 on CCA images, and 2.5±1.0, 3.9±1.1, 2.3±1.1 on bifurcation images. The absolute difference (mean±SD) between each computer-generated contour and the ground truths, taken as the average of three manual delineations, were 1.3±0.8, 3.8±0.8, 5.3±0.5 on CCA images, and 2.3±0.9, 4.6±1.3, 6.3±0.6 on bifurcation images, for the MBQD, IBED and active contour methods, respectively. The MBQD method shows comparable performance with manual delineations on particle images even with poor intima-media layer quality.