Quantitation of the premature infant brain volume from MR images using watershed transform and Bayesian segmentation

Various automated and precise segmentation methods of MR images exist for adult brain, but the segmentation of premature infant brain has been problematic. In this paper, a novel segmentation method for MR images of premature infant brain is proposed. The method utilizes a combination of the watershed transform and Bayesian segmentation techniques. An image of intensity gradients is used as a source for the watershed segmentation method. Watershed basins are then combined according to various criteria to produce a set of approximate segment images that can be used to measure the volume of the premature infant brain. The approximate segmentation is then used as a priori information to help Bayesian segmentation according to the intensity distributions of the gray matter, white matter and cerebrospinal fluid segments of the brain. The method is compared to a standard segmentation method developed for the brain. The comparison is done for both adult and premature infant brain images.