Robust and scalable interactive freeform modeling of high definition medical images

Whole-body anatomically correct high-resolution 3D medical images are instrumental for physical simulations. Unfortunately, only a limited number of acquired datasets are available and the scope of possible applications is limited by the patient's posture. In this paper, we propose an extension of the interactive cage-based deformation pipeline VoxMorph [1], for labeled voxel grids allowing to efficiently explore the space of plausible poses while preserving the tissues' internal structure. We propose 3 main contributions to overcome the limitations of this pipeline: (i) we improve its robustness by proposing a deformation diffusion scheme, (ii) we improve its accuracy by proposing a new error-metric for the refinement process of the motion adaptive structure, (iii) we improve its scalability by proposing an out-of-core implementation. Our method is easy to use for novice users, robust and scales up to 3D images that do not fit in memory, while offering limited distortion and mass loss. We evaluate our approach on postured whole-body segmented images and present an electro-magnetic wave exposure study for human-waves interaction simulations.