The Skull Stripping Problem in MRI Solved by a Single 3D Watershed Transform

A robust method for the removal of non-cerebral tissue in T1-weighted magnetic resonance (MR) brain images is presented. This procedure, often referred to as skull stripping, is an important step in neuroimaging. Our novel approach consists of a single morphological operation, namely a modified three-dimensional fast watershed transform that is perfectly suited to locate the brain, including the cerebellum and the spinal cord. The main advantages of our method lie in its simplicity and robustness. It is simple since neither preprocessing of the MRI data nor contour refinement is required. Furthermore, the skull stripping solely relies on one basic anatomical fact, i.e. the three-dimensional connectivity of white matter. As long as this feature is observed in the image data, a robust segmentation can be guaranteed independently from image orientation and slicing, even in presence of severe intensity non-uniformity and noise. For that purpose, the watershed algorithm has been modified by the concept of pre-flooding, which helps to prevent oversegmentation, depending on a single parameter. The automatic selection of the optimal parameter as well as the applicability are discussed based on the results of phantom and clinical brain studies.