Modelling and improvement of proton computed tomography

Proton computed tomography (pCT) is a promising imaging technique to substitute x-ray CT for more accurate proton therapy treatment planning as it allows to calculate directly proton relative stopping power (RSP) from proton energy loss measurements. A novel pCT scanner (phase II pCT scanner prototype) was completed with a siliconbased particle tracking system and a 5 stage scintillating energy detector. In parallel, a modular software platform was developed to characterize the performance of the pCT system. The modular pCT software platform consists of (1) a Geant4-based simulation modelling the Loma Linda (California, USA) proton therapy beam line and the phase II pCT scanner prototype, (2) water equivalent path length calibration and (3) conversion of the scintillating energy detector, and (4) image reconstruction algorithm for the reconstruction of the RSP of the scanned object. The platform has been validated with respect to experimental measurements and proved to be a valid tool to characterize and optimize the novel pCT system. The results show that the pCT software platform accurately reproduces the performance of the existing phase II pCT scanner prototype with a RSP agreement between experimental and simulated values to better than 1.5%. The pCT software platform was also used to perform a dosimetric evaluation of the phase II pCT scanner prototype. The results are very promising because the dose delivered during a pCT scan was calculated to be 10 time less than the dose delivered during a cone-beam CT scan. Finally, the accuracy of the most likely path calculation in homogeneous and heterogeneous medium was also investigated using a pixelated Medipix detector. The detector was successfully integrated with the experimental phase II pCT scanner prototype. A Geant4 simulation of the pCT-Medipix system was also developed and theoretically predicted, simulated and experimental data were compared and analysed. The agreement between experimental and simulated results is always within one standard deviation and the correlation coefficients between predicted and measured data is close to 1, showing a good agreement between predicted and measured data