Patient realignment in magnetic resonance imaging of the head: an algorithm using mathematical morphology for feature...

To develop a technique for automatic patient realignment in magnetic resonance imaging (MRI), it is essential to extract key features automatically from the various slices of the head as accurately as possible. Such features include the brain, the brain stem, the pons, the corpus callosum and the cerebellum. A feature extraction algorithm has been developed which is based on thresholding a region to a common grey level and then applying mathematical morphology to produce a binary regular region. In addition, a region-filling algorithm has been developed to obtain the complete feature. The scans derived from the T1 spin-lattice relaxation time, which are the fastest of the MRI scans, are used in patient realignment to provide highly textured images. These are difficult to segment using conventional thresholding or edge enhancement techniques due to their 'grainy' appearance, which makes it difficult to isolate key features from the other components found in the slice. We have developed a method for the accurate extraction of the corpus callosum, the cerebellum and the brain area in a sagittal scan of the head. This is carried out by selective thresholding designed to remove the low texture content and then applying morphological techniques.