Abstract
Treatment planning systems for proton therapy require a CT calibration curve relating Hounsfield units to proton stopping powers. An understanding of the accuracy of this curve, together with its limitations, is of utmost importance because the calibration underpins the calculated dose distribution of every patient preparing to undergo proton therapy, independent of delivery technique. The most common approach to the calibration is the stoichiometric method, which is well-defined and, in principle, straightforward to perform. Nevertheless, care must be taken when implementing it in the clinic in order to avoid introducing proton range uncertainties into treatment plans that are larger than the 3.5% that target margins are typically designed to account for. This work presents a variety of aspects related to the user-specific implementation of the stoichiometric calibration, from both a measurement setup and a data-handling point of view, and evaluates the potential impact of each for treatment planning purposes. We demonstrate that two alternative commercial vendors' tissue phantoms yield consistent results, that variable CT slice thickness is unimportant, and that, for a given cross-sectional size, all phantom data can, with today's state-of-the-art beam hardening-related artifact reduction software, be acquired quickly and easily with a single scan, such that the resulting curve describes the calibration well at different positions across the imaging plane. We also show that one should be cautious of using metals in the calibration procedure and of using a single curve for anatomical sites differing widely in size. Further, we suggest that the quality of the parametric fit to the measurement data can be improved by performing a constrained, weighted linear regression. These observations, based on the 40 separate curves that were calculated, should help the medical physicist at any new proton therapy facility in deciding which considerations are worth particular attention.