A hybrid 3D segmentation approach for vasculatures of CTA images

A hybrid 3D segmentation approach for extracting vascular structures of CTA (computed tomographic angiography) scans is presented. In order to remove large amount of granular noises which are commonly exist in CTA images, repetition median filters are employed to smooth the 3D data. Compared to the well performed Vessel Enhancing Diffusion method, the adopted method is more fast, robust, easy to operate, and reproducible. Then, the vascular structure is enhanced by sigmoid filter which converts the vascular intensity distribution to a new level. After that, 3D region growing method is selected to get raw segmentation results. Based on the obtained rough vasculatures, Geodesic Active Contour Level Set algorithm is adopted to refine the segmentation results for its good performance of solving segmentation leakage. Finally, the vascular surface is reconstructed by Marching Cubes and smoothed by Laplacian filter. Experiments show that the developed method can obtain good segmentation vasculature from CTA images.