A new method for nearly automated detection of carotid contours on ultrasound images based on combined region growing and level-set

We developed a new method for "nearly-automated" carotid (CA) contour detection by combining seeded region growing (SRG) and level-set (LS) methods. This technique was applied to cross-sectional vascular images (Aspen, Siemens) acquired during parabolic flight in 5 normal subjects, to detect carotid area changes with gravity (Gz). For each frame, the region resulting from SRG was used to initialize the LS algorithm. This resulted in the final CA contour, from which its area (CAA) was computed. Three consecutive cardiac cycles were selected and max (MCAA) and min CAA (mCAA), as well as their difference D. Compared to 1 Gz, at 0 Gz MCAA was increased by 11.2plusmn5.5% (p=0.07), mCAA was unchanged and D augmented by 44.8plusmn7.2% (p<0.01). The developed method was able to reliably detect the CA lumen compatibly with image quality. Expected changes due to increased stroke volume at 0 Gz were quantified.