%0 Generic
%A Bartelheimer, Kathrin
%C Heidelberg
%D 2020
%F heidok:27738
%R 10.11588/heidok.00027738
%T A Heterogeneous and Multi-Range Soft-Tissue Deformation Model for Applications in Adaptive Radiotherapy
%U https://archiv.ub.uni-heidelberg.de/volltextserver/27738/
%X During fractionated radiotherapy, anatomical changes result in uncertainties in the applied dose distribution. With increasing steepness of applied dose gradients, the relevance of patient deformations increases. Especially in proton therapy, small anatomical changes in the order of millimeters can result in large range uncertainties and therefore in substantial deviations from the planned dose.  To quantify the anatomical changes, deformation models are required. With upcoming MR-guidance, the soft-tissue deformations gain visibility, but so far only few soft-tissue models meeting the requirements of high-precision radiotherapy exist. Most state-of-the-art models either lack anatomical detail or exhibit long computation times.    In this work, a fast soft-tissue deformation model is developed which is capable of considering tissue properties of heterogeneous tissue. The model is based on the chainmail (CM)-concept, which is improved by three basic features. For the first time, rotational degrees of freedom are introduced into the CM-concept to improve the characteristic deformation behavior. A novel concept for handling multiple deformation initiators is developed to cope with global deformation input. And finally, a concept for handling various shapes of deformation input is proposed to provide a high flexibility concerning the design of deformation input.    To demonstrate the model flexibility, it was coupled to a kinematic skeleton model for the head and neck region, which provides anatomically correct deformation input for the bones.   For exemplary patient CTs, the combined model was shown to be capable of generating artificially deformed CT images with realistic appearance. This was achieved for small-range deformations in the order of interfractional deformations, as well as for large-range deformations like an arms-up to arms-down deformation, as can occur between images of different modalities.  The deformation results showed a strong improvement in biofidelity, compared to the original chainmail-concept, as well as compared to clinically used image-based deformation methods. The computation times for the model are in the order of 30 min for single-threaded calculations, by simple code parallelization times in the order of 1 min can be achieved.    Applications that require realistic forward deformations of CT images will benefit from the improved biofidelity of the developed model. Envisioned applications are the generation of plan libraries and virtual phantoms, as well as data augmentation for deep learning approaches.  Due to the low computation times, the model is also well suited for image registration applications. In this context, it will contribute to an improved calculation of accumulated dose, as is required in high-precision adaptive radiotherapy.