From 2D to 3D: Proton Radiography and Proton CT in proton therapy: A simulation study

(1) Purpose
In order to reduce the uncertainty in translation of the X-ray Computed Tomography (CT) image into a map of proton stopping powers (3-4% and even up to 10% in regions containing bones [1-8]), proton radiography is being studied as an alternative imaging technique in proton therapy. We performed Geant4 Monte Carlo simulations for a 2-dimentional (2D) proton radiography system to obtain directly proton stopping powers of the imaged object. In the next step, the object was rotated every 10 degrees to obtain the 3D proton CT, and the iterative reconstruction method was used to reproduce the image.

(2) Materials/methods
In our proton radiography simulation setup we used two ideal (100% efficiency) position sensitive detectors (red squares), with the size of 10x10 cm2 each, to track a single proton entering and exiting a phantom under study. The residual energy of a proton was detected by a BaF2 crystal (yellow cylinder), with a diameter of 15 cm, placed after the second position sensitive detector. A cylindrical phantom with a 2.5 cm diameter and 2.5 cm height was made of CT solid water (Gammex 357, ρ=1.015 g/cm2) and filled with different materials: PMMA (ρ=1.18 g/cm2, red insert), air (ρ=1.21•10-3 g/cm2, below and/or above each inserts), and tissue-like materials: adipose (Gammex 453, ρ=0.92 g/cm2, yellow insert) and cortical bone (Gammex 450, ρ=1.82 g/cm2, blue insert) [9]. The phantom was irradiated with 3x3 cm2 scattered proton beam with an energy of 150 MeV. It was irradiated with 2•105 protons at each of the 36 rotation angles. The phantom was placed perpendicularly to the beam direction allowing a proton to pass through a number of materials with different densities.

(3) Results
First, the energy loss radiographs (a difference between proton beam energy and residual energy deposited in the energy detector) at each of the 36 phantom rotation angles were created. For the iterative reconstruction algorithm, a reference image of the phantom was created in two ways: (1) based on the energy loss in different phantom materials simulated with Geant4, and (2) using a simple back projection algorithm. The reconstruction agrees well with the actual phantom. A maximum of 50 iterations were used showing the smallest mean squared error already after 5 iterations.

(4) Conclusion
First attempt to iteratively reconstruct the cylindrical phantom with more materials on the proton beam shows a satisfactory result. To improve the reconstruction at the material boundaries, additional local iterations will be applied.