Exhibit Hall | Forum 7
Purpose: Dose verification based on proton-induced positron emitters is a promising quality assurance tool. To move a step closer towards practical application, the sensitivity analysis of two factors was investigated: biological washout and depth selection.
Methods: The training dataset was generated based upon a CT image-based phantom and multiple beam energies/pathways, using Monte-Carlo simulation. For the study of biological washout, an analytical model was applied to modify the change of activity profiles over 5 minutes, including both physical decay and biological washout. For the study of depth selection (a challenge linked to multiple-field irradiation), truncations were applied at different lengths (100, 125, 150 mm) to the raw activity profiles. The performances of a worst-case scenario were examined, combining both factors (depth selection: 125 mm, biological washout: 5 mins). The accuracy was evaluated in terms of ∆_range, mean absolute error (MAE) and mean relative errors (MRE over peak region, MRE95).
Results: The machine learning model is found to be more sensitive to depth selection, relative to biological washout. For the worst-case (only activity as input), the mean ∆_range and MRE95 (activity only) are 6.26 mm and 10.1%, respectively. Including anatomical and stopping power prior as auxiliary inputs (activity+HU+SP), is able to significantly improve accuracy/robustness (∆_range : 0.38 mm, MRE95: 3.11%).
Conclusion: The framework shows good immunity to the perturbation associated with the two factors, maintaining good accuracy even in the worst-case. The detection of proton-induced positron emitters, combined with machine learning, has potential to realize online patient-specific verification in proton therapy.
TH- Radiation Dose Measurement Devices: Development (new technology and techniques)