Improving proton therapy planning with photon-counting spectral computed tomography

Abstract: Proton radiation therapy is an alternative to conventional photon radiation therapy, which accounts for the majority of radiation treatments today. The rationale for using protons in radiation therapy lies in their dose deposition properties; photons deposit a radiation dose inversely proportional to the energy, and therefore tissue depth, while protons exhibit a sharp Bragg peak when traversing matter. This property could increase the precision of dose delivery to the target region, and spare healthy tissue in distal and proximal regions. As part of the proton therapy treatment planning, a computed tomography (CT) scan of the patient is performed and the stopping power ratios (SPR) relative to water of the tissues are derived from the CT numbers. Estimates of SPR values are known to be a significant source of uncertainty, leading to increased margins and radiation to healthy tissue. Photon-counting detectors within CT have demonstrated many advantages over their energy-integrating counterparts, such as improved spectral imaging, higher resolution and filtering of electronic noise. In this study, the potential of photon-counting computed tomography for improving proton therapy planning was assessed by training a deep neural network on a domain transform from photon-counting CT images to SPR maps. Since one of the main types of cancer treated with proton therapy are tumours in the brain and head area, head phantoms were constructed and used to simulate photon-counting CT images, as well as to calculate the ground truth SPR value in each image point. The CT images and SPR maps were then used as input and labels to a neural network. Prediction of SPR with this method yielded relative errors of 0.52 - 0.96 %, and RMSE of 0.54 - 1.25 %, which is comparable to methods based on dual energy CT (DECT) using energy-integrating detectors.