Dual‐energy CT for proton therapy: Impact of advanced slice‐wise patient‐thickness estimation methods for improved stopping‐power prediction

Background The direct prediction of stopping‐power ratio (SPR) from dual‐energy CT (DECT) has become gold‐standard in proton therapy. Remaining uncertainties due to patient‐size‐specific CT number variations are mitigated by calibration factors based on patient size defined as water‐equivalent thickness. Purpose To improve SPR prediction, two slice‐wise thickness estimation methods (TEM‐B1 and ‐B2) were compared with the previously used one (TEM‐A). Methods TEM‐A is using the maximum attenuation projections in x ‐ and y ‐direction, while TEM‐B1 and ‐B2 incorporate all voxels of the object to better describe non‐elliptical geometries. Simplified geometries were used to investigate TEM dependencies on several parameters (e.g., object shape, rotation). TEMs were then applied to DECT scans of cylindrical acrylic phantoms with varying diameters and to patient data. Clinical treatment plans were recalculated on generated TEM‐specific SPR datasets, and the impact of different estimated thicknesses on SPR was assessed. Results In contrast to TEM‐A, TEM‐B1 and ‐B2 demonstrated robustness to object shape and rotation. Couch attenuation affected all evaluated TEMs with TEM‐A being most affected. For patient scans, TEM‐B1 and B2 agreed closely but differed from TEM‐A, especially in high diameters. In obese patients, this leads to relative proton range deviations up to 0.3% when comparing TEM‐B1 and TEM‐A. In the sensitivity analysis, TEM‐B1 and ‐B2 maintained SPR uncertainties below ±3% even for cortical bone. Conclusions TEM‐B1 and ‐B2 reduced deviations in thickness estimation and increased robustness to object shape, overcoming the limitations of TEM‐A and improving SPR prediction accuracy.