Assessing accuracy and artefacts in proton stopping power ratio images across four computed tomography imaging workflows using a head-sized electron density phantom

The efficacy of proton beam therapy is limited by stopping power ratio (SPR) prediction uncertainties in patient tissues. This study compared image artefacts and SPR prediction accuracy across a single-energy computed tomography (SECT) and three dual-energy computed tomography (DECT) workflows: SECT with a clinical Hounsfield look-up table (HLUT), two commercial DECT algorithms (DirectSPR and MMSim), and an in-house developed model applied to material density (MD) images, called MD-SPR. SPR images of a head-sized phantom with 24 inserts of tissue surrogate and non-tissue materials were evaluated for image artefacts and compared with measured reference SPRs of the inserts. The root-mean-square SPR differences for tissue surrogates were 0.011 (HLUT), 0.005 (DirectSPR), 0.007 (MMSim), and 0.005 (MD-SPR). For non-tissue materials, the differences were 0.167, 0.028, 0.034, and 0.011, respectively. These results indicate that DECT-based SPR prediction workflows, particularly MD-SPR, can reduce both image artefacts and range uncertainties, compared with a SECT-based HLUT workflow.