EPID-Based 3D In-Vivo Dose Verification Using Deep Learning Inference Modelling
Presentations
PO-GePV-T-217 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: External beam radiotherapy can deliver a highly conformal dose to targets, with steep dose falloff to organs-at risk. Nonetheless, the actual delivered dose distribution will differ from planned dose without in-vivo dose verification during treatment. Electronic portal imaging devices (EPIDs) have been used as a dosimetric tool for pre-treatment and in-vivo patient-specific verification. In this study, a deep learning-based method has been developed to reconstruct in-vivo dose using EPID signals.
Methods: A deep learning-based reconstruction algorithm was utilized to learn a mapping from an EPID dose signal simulated by Monte Carlo method to in-vivo dose distribution at different distance from EPID. A ResNet architecture was trained to reproduce patient dose distribution from the input consisting of a CT reconstruction and a simulated EPID dose signal. Finally, the in-vivo dose distribution was compared with the planned dose to evaluate the quality of prediction.
Results: Mean absolute percentage error (MAPE) and a gamma analysis were quantified between the in-vivo dose distribution and the planned dose. Both comparisons indicate that the reconstructed in-vivo dose was in a good agreement with the original dose.
Conclusion: EPID-based 3D in vivo dosimetry reconstruction provides an efficient approach to monitor the accuracy of dose delivery during treatment.
Keywords
Electronic Portal Imaging, In Vivo Dosimetry, 3D
Taxonomy
TH- External Beam- Photons: portal dosimetry, in-vivo dosimetry and dose reconstruction
Contact Email