Which automated ultrasonic boundary detection algorithm provides the most accurate quantification of human carotid artery intima-media thickness compared to manual measurements?
A dynamic programming algorithm provides highly accurate automated quantification of carotid intima-media thickness from ultrasound images, closely correlating with manual measurements.
In this paper we examine four algorithms for automated ultrasonic boundary detection, and describe the application of these algorithms to the quantification of the intima-media thickness (IMT) in the human carotid artery. The first algorithm uses a dynamic programming approach to identify the boundary that minimizes a certain cost function. The second algorithm is based on finding points of maximum gradient. The third algorithm employs a mathematical model describing the intensity profile perpendicular to the two boundaries defining the IMT. The last algorithm is based on defining a template representing the intensity profile across boundary and applying a matched filter procedure to find the image region that best matches it. The authors also present a quantitative and qualitative comparison between the four algorithms examined. It is shown that the dynamic programming algorithm provides superior performance in terms of accuracy and robustness. The correlation coefficients between automated measurements and manually obtained reference values were 0.96, 0.94, 0.63, and 0.85 for the dynamic programming, the maximum gradient the model-based, and the matched filter algorithm, respectively (n=30).
Gustavsson et al. (Wed,) studied this question.