Room: Exhibit Hall | Forum 6
Purpose: Online range/dose verification based on measurements of proton-induced positron emitters is a promising strategy for quality assurance in proton therapy. Because of the nonlinear correlation between the dose and the activity profiles in addition to the adverse influence of noise, the relationship between these two needs to be established using machine learning approaches. Five recursive neural network based (RNN) structures were investigated.
Methods: Simulations were carried out with a spot-scanning proton system using GATE-8.0 and Geant4-10.3 toolkit with a CT-based patient phantom. The 1D distributions of positron emitters and radiation dose were obtained. Training data (2000 in total, 80% for training and 20% for testing) were generated for different beam energy, irradiation positions and counting statistics. The absolute range shift (MSE) and dose uncertainty in terms of mean square error (LOSS, LOSS50 and LOSS20) were quantitatively studied.
Results: The feasibility of proton range and dose verification using the RNN-based framework was demonstrated. All models are able to converge after 150 epochs without overfitting. When the signal-to-noise ratio (SNR) of the activity profiles is about 4.5 (after 8 proton pulses), the MSE is within 1 mm and the overall dose accuracy is within <5%. The accuracy of prediction is found to be inferior for large energy scenarios, compared to low and medium energy. BiGRU demonstrates the most stable and accurate performance with the presence of noise for both range and dose verification.
Conclusion: Our proposed framework may enable highly accurate online range and dose verification, due to its strength in identifying the correlation between the dose profile and positron emitters, taking into the context among different depths. The machine learning-based framework promises to provide a reliable and effective way for online monitoring, and ultimately allows for adaptive proton therapy.
Not Applicable / None Entered.