Segmentation of Magnetic Resonance Brain Images Using Watershed Algorithm

An important area of current research is obtaining more information about brain structure and function. Brain tissue is particularly complex structure and its segmentation is an important step for studies intemporal change, detection of morphology as well as visualization in surgical planning, volume estimation of objects of interest, and more could benefit enormously from segmentation. Magnetic resonance imaging (MRI) is a noninvasive method for producing tomographic images of the human brain. Its Segmentation is problematic due to radio frequency inhomogeneity, caused by inaccuracies in the magnetic resonance scanner and by movement of the patient which produce intensity variation over the image, and that makes every segmentation method fail. The aim of this work is the development of a segmentation technique for efficient and accurate segmentation of MR brain images. The proposed technique based on the watershed algorithm, which is applied to the gradient magnitude of the MRI data. The watershed segmentation algorithm is a very powerful segmentation tool, but it also has difficulty in segmenting MR images due to noise and shading effect present. The known drawback of the watershed algorithm, over-segmentation, is strongly reduced by making the system interactive (semi-automatic), by placing markers manually in the region of interest which is the brain as well as in the background. The background markers are needed to define the external contours of the brain. The final part of the segmentation takes place once the gradient magnitudes of the MRI data are calculated and markers have been obtained from each region. Catchment’s basins originate from each of the markers, resulting in a common line of separation between brain and surrounding. The proposed segmentation technique is tested and evaluated on brain images taken from brainweb. Brainweb is maintained by the Brain Imaging Center at the Montreal Neurological Institute. The images had a combination of noise and intensity non-uniformity (INU). By making the system semi-automatic, a good segmentation result was obtained under all the conditions (different noise levels and intensity non uniformity). It is also proven that the placement of internal and external markers into regions of interest (i.e. making the system interactive) can easily cope with the over-segmentation problem of the watershed.

Preview PDF 1000548980_t_FK_2004_32.pdf Download (195kB)

Actions (login required)

View Item