Detection and quantitative parameterization of brain blood vessels in magnetic resonance images (MRI) are an important aid to diagnosing neoplasmic diseases, planning surgical operations or detecting the atrophy of blood vessels. Fast and effective computer programs are needed to extract quantitative information from MRI data - to increase objectivity, accuracy and repeatability of the diagnosis. To develop such programs we must use algorithms for 3D images segmentation, necessary to build geometrical models of the blood vessels. These models are then used for vessel tree visualization and quantitative description (Fig. 1). [GRAPHICS] There are few known algorithms for detection of blood vessels in MR images. One of them is level set algorithm [1]. The paradigm of the level set is that it is a numerical method for tracking the evolution of contours and surfaces. Instead of manipulating the contour directly, the contour is embedded as the zero level set of a higher dimensional function called level-set function. Another useful method is based on so called vesselness function, derived from eigenvalues of the Hessian matrix [2]. Both algorithms feature long computation time and need interaction to indicate seed points from which the vessel tracking starts. Use of mathematical morphology methods combined with region growing algorithm is proposed for vessel segmentation in this paper [3]. The developed algorithm is much faster and doesn't require user intervention. Real and synthesized MR images were used to test the properties of the algorithm. Real MRI data of brain contains blood vessels brighter or similar to other neighboring tissues. Tissues situated near the skull also have brightness similar to the vessels. Another tested data were numeric phantoms with added noise. Those phantoms represents bright vessels with known diameter, length and random orientation in space on darker background. The goal was to detect areas with blood vessels and to create binary image of them. That image was a source to perform visualization of detected vessels. Phantom images were used to estimate volume of cylinders simulating vessels and to compare results with real volume of simulated objects. An algorithm was implemented in ITK/VTK [4] which is an open source library which is a collection of procedures useful in 3D medical image analysis. ITK library was used to perform segmentation of blood vessels. A novel sequence of mathematical morphology algorithms was developed to extract the regions in 3D magnetic resonance data that correspond to blood vessels. A procedure for automatic image intensity threshold selection that allows to eliminate near skull tissues regions (irrelevant to the blood vessels) was also developed. VTK library was used to visualize results of the vessel tree extraction and modeling, because this library has a very efficient graphics rendering engine based on Open GL [5]. Good visualization was obtained with the use of marching cubes algorithm [6] and surface smoothing. The volume of blood vessels phantoms estimated on the segmentation results was always larger than the actual volume. The volume estimation error decreases with the vessel diameter. The time of vessel tree extraction from 279x213x132 voxel images was 3-4 times shorter compared to vessel tracking with the use of Frangi filter (Hessian matrix) [7]. Further work is planned to improve the accuracy of vessel segmentation with the proposed algorithm through the use of adaptive thresholding instead of the global one.
