An integrated system for the complete segmentation of the common carotid artery bifurcation in ultrasound images

Abstract

The complete segmentation of the common carotid artery (CCA) bifurcation in ultrasound images is important for the evaluation of atherosclerosis disease and the quantification of the risk of stroke. The current research work further evaluates and validates a semi-automated (SA) snake's based segmentation system suitable for the complete segmentation of the CCA bifurcation in two-dimensional (2D) ultrasound images. The proposed system semi-automatically estimates the intima-media thickness (IMT), the atherosclerotic carotid plaque borders and dimensions, the internal carotid artery (ICA) origin's stenosis, the carotid diameter (D), as well as other geometric measurements of the atherosclerotic carotid plaque. The system was evaluated on 300 2D longitudinal ultrasound images of the CCA bifurcation with manual (M) segmentations available from a neurovascular expert. No statistical significant differences between all M and SA IMT, plaque and D segmentation measurements were found. In a future study, texture features extracted from the intima-media complex (IMC) may be used to separate subjects in high and low risk groups, which may develop a stroke. However, a larger scale study is required for evaluating the system before its application in the real clinical practice.

Two Automated Techniques for Carotid Lumen Diameter Measurement: Regional versus Boundary Approaches

“…However, we have found that some authors pointed out the need for automated measurement of carotid lumen diameter [47][48][49]. The lumen segmentation algorithms have been classified into regional-based vs. boundary-based: region-based [17,18], boundary-based [19][20][21][22]. Here, we describe each of these techniques and their limitations and discuss how they validated the measured parameters.…”

“…The study by Loizou et al [20] was concentrated on bifurcation region segmentation and diameter estimation through speckle reduction filtering and snake-based segmentation. This algorithm faced the challenge of initialization of snakes and did not give a high value of coefficient of correlation between their method and manual reading (CC = 0.63).…”

“…We have classified these techniques into three categories [10,15,16] as: region-based [17,18], boundary-based [19][20][21][22] or fusion-based [6].…”

Frequency-domain active contours solution to evaluate intima–media thickness of the common carotid artery

“…For example, in [37], there are no LD error statistics, only an overlap index of 79%. In [42], the LD error is under 100 m on average, but there is no data about statistical dispersion (see Table 6).…”

“…These proposals implement two stages, one to initialize the contours and a second one to segment the artery walls. Following the same idea, other authors have proposed methods based on models [36] or different implementations of active contours [37,38], like active contours without edges [39] based on the Chan-Vese snake [40], or frequency-formulation of active contours [41]. The latter, needs to perform twice the active contour segmentation,increasing the computational cost, and this work lacks the comparison of the results with a ground truth segmentation.…”