Segmentation Of Vascular Tree In The Medical Images Based On Multiscale Hessian Enhancement And Active Contour Method

Vascular diseases are among the most important public health problem in developedcountries. Meanwhile, Cardiovascular and Cerebrovascular Diseases (CVDs) is currentlyone of the leading causes of morbidity and mortality worldwide. For various medical diag-nostic tasks, it is necessary to measure the vessel width, reflectivity, tortuosity and abnormalbranching, which are of importance to diagnose the severity of vascular disease and de-termine the treatment therapy. In general, vessel segmentation and visualization is indeed afundamental step for computer-aided diagnosis, treatment, surgical planning and navigation.Various segmentation methods have been proposed for vascular structure varying dependingon the imaging modality, application domain and other specific factors. There is no singlesegmentation method can extract vasculature from every medical image modality.This thesis takes the abdomen magnetic resonance angiography (MRA) image adjacentliver and other organs as the segmentation object and reviews the main segmentation meth-ods. We proposed an automatic vascular segmentation method, which combines Hessian-based multiscale filtering and a modified level set method. This method mainly containsvessel enhancement and vessel segmentation. Firstly, the morphological Top-Hat transfor-mation is adopted to attenuate background. Then Hessian-based multiscale filtering is usedto enhance vascular structures by combining Hessian matrix with Gaussian convolution totune the filtering response to the specific scales. Because Gaussian convolution tends to blurvessel boundaries, which makes scale selection inaccurate, an improved level set method isfinally proposed to extract vascular structures by introducing an external constrained termrelated to the standard deviation σ of Gaussian function into the traditional level set.To evaluate our segmentation method, we tested it on synthetic images with vascular-like structures and real abdomen MRA images, and also compared with other methods.Experimental results demonstrate that our approach could extract vascular structureseffectively without any human intervention, and is insensitive to adjacent organs.