10/2 Predictive properties of statistical shape models

Originally, statistical shape models were dominantly applied to constrain the result of a segmentation process to an anatomically meaningful form. In many applications today the main goal of using the prior information acquired during the training phase is the prediction of a patient-specific organ shape when only partial information is given. A typical example in computer assisted orthopaedic surgery is the possibly precise estimation of a 3D bone model from X-ray projection(s) or from manually digitised points on a very limited, surgically reasonably accessible patch of its surface.

The goal of this work package is to systematically investigate the predictive properties of statistical bone shape models, to characterise both the predictors and the local remaining variability of interpolated shape in a quantitative manner. We will develop tools for the systematic identification of optimal form predictors and the search for a statistical framework allowing the balanced use of interpolation, based on the shape model and the actually acquired intra-operative information. New statistical shape modelling techniques will be studied. In addition, the characterisation of shape dependences (coupled shape models) between structures of the skeleton will be investigated.