Proceedings Article | 13 March 2013
KEYWORDS: Image segmentation, Arteries, Laser induced breakdown spectroscopy, 3D image processing, Ultrasonography, Independent component analysis, Simulation of CCA and DLA aggregates, Image processing algorithms and systems, Visualization, Speckle
Carotid atherosclerosis is a major cause of stroke. Imaging and monitoring plaque progression in 3D can better classify
disease severity and potentially identify plaque vulnerability to rupture. In this study we propose to validate a new
semiautomatic carotid lumen segmentation algorithm based on 3D ultrasound imaging that is designed to work in the
presence of poor boundary contrast and complex 3D lumen geometries. Our algorithm uses a distance regularized level
set evolution with a novel initialization and stopping criteria to localize the lumen-intima boundary (LIB). The external
energy used in the level set method is a combination of region-based and edge-based energy. Initialization of LIB
segmentation is first done in the longitudinal slice where the geometry of the carotid bifurcation is best visualized and
then reconstructed in the cross sectional slice to guide the 3D initialization. Manual initialization of the contour is done
only on the starting slice of the common carotid, bifurcation, and internal and external carotid arteries. Initialization of the
other slices is done by eroding segmentation of previous slices. The user also initializes the boundary points for every
slice. A combination of changes in the modified Hausdorff distance (MHD) between contours at successive iterations
and a stopping boundary formed from initial boundary points is used as a stopping criterion to avoid over- or under-segmentation.
The proposed algorithm is evaluated against manually segmented boundaries by calculating dice
similarity coefficient (DSC), HD and MHD in the common carotid (C), carotid bulb (B) and internal carotid (I) regions
to get a better understanding of accuracy?. Results from five subjects with <50% carotid stenosis showed good
agreement with manual segmentation; between the semiautomatic algorithm and manuals: DSC (C: 86.49± 9.38, B:
82.21±8.49, I: 78.96±7.55), MHD (C: 3.79 ± 1.64, B: 4.09 ± 1.71, I: 4.12 ± 2.01), HD (C: 8.07±2.59, B: 10.09±3.95, I:
11.28±5.06); and inter observers: DSC (C: 88.31±5, B: 82.45±7.57, I: 82.03±8.83), MHD (C: 3.77±2.09, B: 4.32±1.88, I:
4.56±2.24), HD (C: 7.61±2.67, B: 10.22±4.30, I: 10.63±4.94). This method is a first step towards achieving full 3D
characterization of plaque progression, and is currently being evaluated in a longitudinal study of asymptomatic carotid
stenosis.
