A method is proposed for segmenting a 2- or 3-D space, spanned by a set of medical data comprising intensity values at locations within the space, to estimate the position of an object of medical significance. The method using a level set function having a level set which provides a model of boundary of the object. The level set function is iteratively updated by a force defined based on the medical data. For computational efficiency, only the force in the narrowband of the model boundary needs to be calculated. Minimization of an energy function, related to the force, provides a termination condition for the iteration. High level knowledge can be incorporated in several ways, such as by an explicit force term which takes over from the force based on the medical data when prior knowledge of the object is about to be violated by the model.