Extraction Of Blood Vessels From Medical Images

Automatic extraction of blood vessels for medical images plays an important role in medical image processing. A clear segmentation of blood vessels could assist surgeon and specialist to perform better treatment diagnosis. In this thesis, extraction of blood vessels both from retinal images (2D) and liver CT images (3D) have been studied and evaluated.A lot of anatomical information associated with physiological changes can be found in retinal images. This makes it possible to detect many diseases that are relevant to cardiovascular in early stages via segmenting and visualizing the retinal blood vessels. In this work, an automated method is proposed to extract blood vessels from retinal images, which consist of the two image analysis steps. Based on the observation that cross-section intensity profiles of blood vessels are Gaussian-shaped, an input retinal image is first smoothed by a set of Gaussian filters along different orientations. Eventually a so-called matched filter response (MFR) image is created for identifying the seeds of the segmentations. Feature points or seeds are then extracted from the MFR image as starting points for improved iterative adaptive local threshold probing. Experimental results have demonstrated that the underlying method is very powerful when detecting blood vessels for retinal images, in particular for small vessels.In liver surgery, one of main challenges is to conduct the surgical planning first before the treatment, which is hugely time-consuming for manual estimation and segmentation and therefore is tempted to create human-errors. Understanding the structure of hepatic vessels is of most importance in order to achieve efficient and clear-cut planning for precision. In this presentation, a novel thresholding based segmentation algorithm has been exploited and applied to extract liver vessels, which incorporates both the intensity and gradient information of liver CT images. The threshold for vessel segmentations is adaptively selected by Robust Automatic Threshold Selection (RATS). At first, Gaussian filters and median filters must be performed over the entire set of raw CT images, for achieving a vessel enhancement for better segmentation. Then the following procedures can be performed over those pre-processed images achieved by the two filters: pre-segmentation of liver vessels with a thresholding based method, estimation of an adaptive intensity threshold T, vessel segmentation according to T and analysis of the structure of liver vessels. The method is evaluated experimentally on a set of volume CT data. Experimental results show that the proposed segmentation algorithm is very robust and fast for liver vessel extractions.