MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Deep Convolutional Neural Network for Scatter Prediction in Lung Proton SBRT Treatment

SM Gadoue

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: To predict the scatter correction factor in passively scattered proton beam MU modeling forlung SBRT patients.
Methods: The scatter conditions depend on various parameters such as beam energy, field size,
chest wall thickness, and the amount of lung tissue within the treatment field. Geant4 Monte Carlo simulation toolkit has been used to calculate scatter correction factors for patient setups of varying field sizes, thickness of chest wall and lung tissue traversed by proton beam. The obtained factors are then introduced along with their corresponding parameters to the widely used U-net architecture
for training. 70% of the dataset were randomly selected as the training set, while 10% are chosen for validation, and the remaining 20% for testing.
Results: The results indicate that U-net can accurately predict the scatter correction factor for lung
SBRT proton fields. Table 1 displays the relative difference between the simulated and predicted correction factor for different scenarios. Overall, the percentage difference between the values is below 1% for all cases, with the highest error associated with larger lung tissue thickness.
Conclusion: Using U-net it was possible to predict the correction required for the calculation of the dose/MU in SBRT lung lesions. The deep learning-based prediction provides an opportunity to
improve the workflow in the clinic, and reduces the amount of time spent performing repetitive
simulation and/or measurements by physicists.

Keywords

Protons

Taxonomy

Not Applicable / None Entered.

Contact Email