We present an improved computational method to optimize the calibration curves to convert the Hounsfield Unit (HU) of the X-ray planning computed tomography (CT) to proton relative stopping powers (RSP) for patient-specific proton therapy treatment planning. The commonly used optimization method (Schneider et al. 2005 MP 32, Doolan et al. 2015 PMB 60) calculates a proton digitally reconstructed radiography (pDRR) by converting the HU of the X-ray planning CT into RSP and integrating along the line of beam penetration. The deviation of the pDRR from a measured proton radiography is minimized by optimizing the parametrization of the HU-RSP curve. This scenario assumes that all discrepancies between proton radiography and pDRR originate from uncertainties in the HU-RSP curve which is not satisfied if the pDRR is obtained by a simple projection along straight lines. Instead, multiple Coulomb scattering (MCS) and, in case of an active scanned beam delivery system, the finitebeam profile lead to an imperfect representation of geometric structures (“blurring”) in the measured proton radiography.We analyze these effects and propose to accurately model them to obtain an extended optimization method. We furthermore perform a thorough statistical treatment and propose an analytical formulation to determine the gained accuracy of the optimized HU-RSP curve. We demonstrate that without extending the optimization scheme, spatial blurring in the proton radiographies can cause up to 10% deviation between the optimized and the ground truth HURSP calibration curve. Instead, results obtained with our extended method reach 1% or better correspondence.Our contribution underlines the potential of a single proton radiography to generate a patientspecific calibration curve and to improve dose delivery by optimizing the HU-RSP