Abstract
Recently, in-beam positron emission tomography (PET) has been actively researched for reducing biological washout effects and dose monitoring during irradiation. However, the positron distribution does not precisely reflect the dose distribution since positron production and ionization are completely different physical processes. Thus, a novel in-beam system was proposed to determine proton dose range by measuring scattered protons with dozens of scintillation detectors surrounding the body surface. While previous studies conducted a preliminary experiment with a simple phantom, we simulated more complex situations in this paper. Especially, we conducted three stepwise simulation studies to demonstrate the feasibility of the proposed method. First, a simple rectangular phantom was reproduced on simulation and irradiated with protons for obtaining current values and Monte Carlo (MC) dose. Next, we trained a deep learning model to estimate 2-dimensional-dose range (2D-DL dose) from measured current values for simulation (A). We simulated plastic scintillators as detectors to measure the scattered protons. Second, a rectangular phantom with an air layer was used, and 3D-DL dose was estimated in simulation (B). Finally, a cylindrical phantom that mimics the human body was used for confirming the estimation quality of the simulation (C). Consequently, the position of the Bragg peak was estimated with an error of 1.0 mm in simulation (A). In addition, the position of the air layer, as well as the verifying peak position with an error of 2.1 mm, was successfully estimated in simulation (B). Although the estimation error of the peak position was 12.6 mm in simulation (C), the quality was successfully further improved to 9.3 mm by incorporating the mass density distribution obtained from the computed tomography (CT). These simulation results demonstrated the potential of the as-proposed verification system. Additionally, the effectiveness of CT utilization for estimating the DL dose was also indicated.