Fusion of edge enhancing algorithms for atherosclerotic carotid wall contour detection in computed tomography angiography

The aim of this study is to assess the feasibility and performances of the fusion of edge enhancers in in vivo computed tomography angiography (CTA) images for automatic segmentation of outer and inner vessel walls, in presence of atherosclerotic plaques. From 4 patients' CTA exams (stenosis degrees 70% - 95%) the slices representing plaques were extracted (223 images) and hand segmented by a trained operator for the vessel walls. The analyzed slices depict the common and internal carotid arteries and the carotid bifurcation. The automatic protocol exploits two different categories of image edge enhancers: 5 edge detectors (Sobel, Prewitt, Roberts, laplacian of gaussian (LOG) and Canny) and 5 filters/mapping functions (laplacian filter, gradient map (GM), Otsu thresholding (aT), local range map (LRM) and standard deviation (STD) map). The mean correlation coefficient between the manual and the automatic masks is 48% [17%, 64%]. By selecting the GM, LRM and STD algorithms only, the mean performance is improved up to 58%. This methodology was proven to be comparable to the manual one. The correct selection of the edge enhancers is critical for the performance optimization: GM, LRM and STD showed to be the most suitable for our purpose.