A hierarchical approach to model-based skull segmentation in MRI volumes

We present a model-based approach for segmentation of the skull from T1 weighted MR images of the human head. Segmentation is performed by fitting a statistical shape model of the skull into a crudely pre-segmented version of the image. This yields a segmentation result that is constrained to the normal skull anatomy and thus gives a statistically meaningful approximation of the skull structure, even in places where bone cannot be distinguished from the surrounding tissue. We propose a multi-resolution approach to model fitting and show how a hierarchy of shape models can be used to increase the flexibility of the model. To validate our method, we show fitting results for different MR images and perform a quantitative comparison with a data set where a ground truth segmentation from a CT image is available. Our experiments confirm that even such difficult segmentation tasks as skull segmentation from MR images become feasible given a strong shape prior.