Interactive image segmentation tools for building models of 3D medical images
Computer-aided interpretation of 3D medical images (eg magnetic resonance images) is greatly assisted by the use of statistical models of shape. Such models consist of mathematical parameterisations of the surfaces of organs and their interior structure. If these parameters can be derived from the statistics of a set of training images they can provide very powerful descriptions for use in image segmentation (finding organ surfaces in new, unseen images) or making comparisons between images (such as disease and control groups).Recent research has developed automatic methods of parameterising the shapes and variation in shape of the surfaces. However, the issue remains of obtaining appropriate surfaces from the training images in a consistent fashion. Currently this problem is addressed heuristically, and surfaces are frequently defined by tedious and time-consuming manual annotation (interactive segmentation). While tools exist for assisting in the interactive segmentations, there is a need for a generic framework within which a range of segmentation approaches can be driven under user control to make this process much more efficient and convenient for the user (usually an expert radiologist).
This project will seek to develop such a framework by extending efficient interactive segmentation methods that have recently appeared in the literature to deal with the case of interest here in which a collection of similar images is to be segmented. Among other things, the methods should be able to 'learn' the properties of the images to minimise the amount of interaction required over the set. The research will make use of existing 3D image sets of the brain and other anatomical structures. Interactive segmentation methods have been used previously in media-oriented applications. Here we seek to apply them to more challenging medical images. In addition to extending existing methods, the student will develop new techniques. As user interaction is central to this approach it will also be necessary to take into account human interface issues, such as visualisation of surfaces on 3D images. The student will be able to take advantage on considerable experience in this area within the research division.
