Optimising proton stopping power ratio prediction with spectral cone-beam CT

Objective. Cone-beam computed tomography (CBCT) is used for patient positioning in proton therapy, but not directly for treatment planning due to its inferior image quality compared to fan-beam CT. One way to improve its value for proton radiotherapy might be to use CBCT setups capable of extracting spectral information, which can be realised through several hardware configurations. Here, we compare different setups w.r.t. to their capability of predicting proton stopping power ratios (SPRs).Approach. We investigate six different spectral CBCT realisations in a simulation study, namely a single-source setup with either a dual-layer detector or a photon-counting detector (PCD), a kVp-switching setup with either an energy-integrating detector (EID) or a PCD, and a dual-source setup with either EIDs or PCDs. Our figure of merit is the normalised Cramér-Rao Lower Bound (nCRLB) on SPR variance based on projection data. We take (cross)scatter into account, and compare ideal and realistic detector models to help guide future detector developments. Each setup is optimised w.r.t. source spectra, mAs ratios and energy bin settings (where applicable).Main results. Assuming a realistic detector response, setups with a kVp-switching source perform best, with the setup paired with an EID slightly outperforming the PCD-based setup (nCRLBs of 2.74 and 2.81, respectively). However, if the mAs ratio of the kVp-switching source is fixed, the performance of the kVp-switching setup with an EID is significantly degraded (nCRLB = 9.46) and outperformed by PCD-based setups, with nCRLBs of 3.27, 3.45 and 3.60 for the dual-source setup with two PCDs, the single-source setup and the kVp-switching setup with one PCD, respectively. Spectra with higher source voltage or wider spectral separation generally yield lower CRLB values, and avoiding the spectral distortion caused by charge sharing in direct-conversion PCDs promises to lower CRLB values by about a third.Significance. We present an extensive comparison of spectral CBCT setups for their application in proton radiotherapy, using a methodology that allows to compare their theoretical limit of performance without being influenced by the choice of reconstruction algorithm or the conversion scheme from Hounsfield units to SPR values.