Program Information
Neural Network-Based Range Verification for Proton Therapy Based On Prompt Gamma Emissions
H Peng*, L Xing , Stanford Univ School of Medicine, Stanford, CA
Presentations
TU-C1-GePD-J(A)-4 (Tuesday, August 1, 2017) 9:30 AM - 10:00 AM Room: Joint Imaging-Therapy ePoster Lounge - A
Purpose: Online dose monitoring in proton therapy is currently being investigated with prompt-gamma detection, which is correlated with proton range and dose deposition.
Methods: In our study, the features are selected to be the differences of prompt gamma profiles for different proton energies and Bragg peaks. Simulations were carried out with a clinical spot-scanning proton therapy treatment using the Geant4 V9.4p01 toolkit. The proton beams tested were mono-energetic beams at 130 MeV, 125 MeV, 120 MeV and 115 MeV. A water phantom had a dimension of 10x10x30 cm3 and a voxels size of 1x1x1 mm3. The method implemented a neural network classification model comprising 2 layers and 10 neurons. The classifier’s performances were evaluated using confusion matrix and receiver operating characteristic curves (ROCs). In total, 50 data sets were generated for each proton energy (70% training, 30% testing). Two important parameters were evaluated: kernel size and detection efficiency.
Results: Without any kernel and detection model applied, the classifier is able to classify three proton energies/ranges with 100% accuracy indicating their strong correlations with the prompt gamma profiles. No significant difference is found among different kernel sizes ranging between 1 mm and 16 mm, which relives our burden in optimizing the location of sensors as well as collimators in-between sensors in order to maintain spatial information of dose profiles. In addition, the accuracy of the classifier is found to be dependent on the detection efficiency. For a kernel size of 8 mm, the accuracy is found to be ~70% for a detection efficiency of 0.02%.
Conclusion: The feasibility of proton range verification using neural network was proved. The proposed method will allow us to develop a cost-effective online range/dose verification system for proton therapy with a limited number of sensors located at different depths.
Contact Email: