A Framework for Predictive Modeling of Intra-operative Deformations: A Simulation-Based Study

Deformations that occur between pre-operative scans and the intra-operative setup can render pre-operative plans inaccurate or even unusable. It is therefore important to predict such deformations and account for them in pre-operative planning. This paper examines two different, yet related methodologies for this task, both of which collect statistical information from a training set in order to construct a predictive model. The first one examines the modes of co-variation between shape and deformation, and is therefore purely shape-based. The second approach additionally incorporates knowledge about the biomechanical properties of anatomical structures in constructing a predictive model. The two methods are tested on simulated training sets. Preliminary results show average errors of 9% (both methods) for a simulated dataset that had a moderate statistical variation and 36% (first method) and 23% (second method) for a dataset with a large statistical variation. Use of the above methodologies will hopefully lead to better clinical outcome by improving pre-operative plans.