Clinical applicability of the Compton camera for Prompt y-ray Imaging during proton therapy

(1) Purpose
In order to guarantee the best outcome of a therapeutic irradiation with protons and other light ions a non-invasive in-vivo range verification is desired. One approach in this field is Prompt y-ray Imaging (PGI). A possible detection system for the prompt y-rays is the Compton camera. Several groups have been working on the construction of Compton camera prototypes (Kormoll et al., 2011; Llosá et al., 2013; Thirolf et al., 2014; Hueso-González et al., 2014). Up to now, Compton cameras have not been used in clinical practice for the monitoring of particle therapy. Therefore, by means of Geant4 simulations, we performed an end-to-end test to evaluate the clinical applicability of a Compton camera detection system and to determine the requirements regarding hardware and image reconstruction.
(2) Materials/methods
First, a treatment plan for a therapeutic proton irradiation for the head-neck region was prepared using XiO (Electa AB, Sweden). Based on this treatment plan, the y-ray emissions from the patient's tissue were simulated with Geant4. As a next step, the detector response was modelled, also with Geant4, for two large Compton cameras arranged around the patient in an angle of 90 degrees. Large-area detectors were already recommended (McCleskey et al., 2015). Each camera was built up from a scatter layer (CdZnTe) of dimension 10 × 10 × 0.5 cm3 and an absorber layer (LuSiO) of size 20 × 20 × 2 cm3. In practice, these cameras would be replaced by several smaller camera modules. For the simulation of the detector response a total number of previously simulated y-ray emissions were used as input corresponding to an applied dose of 1 Gy, i.e. a common dose of one field of one treatment session. After extracting the resulting coincident events, the image was reconstructed using a 3D MLEM algorithm (Schoene et al., 2015). The impact of the number of events as well as background on the image quality was also studied.
(3) Results
Figure 1 shows the images for the planned dose, the distribution of the y-ray emissions and the reconstructed image obtained with 128 iterations of the MLEM algorithm. For the considered number of events and the chosen voxels of 5 mm3 the runtime of the reconstruction was about two days on a cluster.

(4) Conclusions
For the considered large camera system and the realistic patient scenario with a dose of 1 Gy adequate images are obtained, which certainly could be applied to detect range deviations in the range of centimeters. Thus, this study demonstrates in principle the clinical applicability. The reconstruction algorithm still has potential for improvements with respect to performance. Furthermore, in practice, the costs of this complex detection system could lead to the preference of simpler methods of PGI.