A 2D-3D Deformable Image Registration Framework for Proton Radiographies in Adaptive Radiation Therapy

The role of radiographic ion imaging in Adaptive Radiation Therapy is investigated relying on a 2D-3D Deformable Image Registration (DIR) framework. Monte Carlo simulations of proton radiographies from list-mode and integration-mode data acquisitions are performed based on a clinical dataset. Following validation on a geometrical dataset, the 2D-3D DIR is investigated with respect to the reference 3D- 3D DIR as a function of different number of proton radiographies for both detector technologies. The 2D-3D DIR framework is based on an optimization algorithm that embeds the forward-projection of the iteratively deformed treatment planning CT image, calibrated to relative stopping power (RSP) of ion. By definition, the geometrical dataset requires the geometrical forward-projection to match the (idealized) "ion" radiographies. For the clinical dataset, the geometrical forward- projection along each pencil beam matches the proton radiographies from integration-mode data, which are based on the dominant range component. Instead, proton radiographies from list-mode data require the forward-projection along the most likely proton trajectory. The feasibility of the 2D-3D DIR relying on a reasonably limited but uniformly distributed number of radiographies is demonstrated on the geometrical dataset. Promising results are obtained by a minimum number of two proton radiographies in the clinical dataset for both detector technologies. Limitations are observed due to the low proton statistics, corresponding to a clinically acceptable dose of 1.9 mGy for the total 180 proton radiographies. More challenges are identified in proton radiographies from integration-mode data.