Neural Network Based Event Classification to Improve Compton Imaging for Proton Beam Range Verification
C Barajas1, G Kroiz1, J Polf2*, S Peterson3, D Mackin4, S Beddar4, M Gobbert1, (1) University Of Maryland, Baltimore County, (2) University of Maryland School of Medicine, Baltimore, MD, (3) University of Cape Town, Rondebosch, ,ZA, (4) UT MD Anderson Cancer Center, Houston, TX
Presentations
TU-B-TRACK 6-3 (Tuesday, 7/27/2021) 11:30 AM - 12:30 PM [Eastern Time (GMT-4)]
Purpose: To determine if Compton camera (CC) images of prompt gamma (PG) emission that occurs during proton radiotherapy (RT) can be improved by using a neural network (NN) to identify 1) "True" and “False” PG events, and 2) the correct interaction order for “True” PG events recorded by the CC.
Methods: The Monte Carlo plus Detector Effects (MCDE) model was used to produce realistic PG interactions that occur in a CC during clinical proton RT beam delivery. The PG data was used to train a fully connected NN to identify True (due to a single PG) and False (due to multiple PGs) PG double scatter (DS) and triple scatter (TS) events. The NN predicted type and interaction order for the events were compared to the known values (from the MCDE model) to determine the accuracy of the NN predictions. Images were reconstructed using the raw MCDE data and a dataset with the NN identified False DS/TS events removed and using the NN predicted True DS/TS interaction order.
Results: The fully trained NN was able to predict the PG event type (True/False) and interaction order with > 80% accuracy. The proton beam range could easily be identified on the PG images reconstructed using NN processed data, even for beams delivered at the highest clinical dose rates.
Conclusion: NN processing of CC data can significantly improve the acquired images of PG emission occurring during the delivery of clinical proton RT beams. These results show that further studies of NNs for this purpose are warranted.
Funding Support, Disclosures, and Conflict of Interest: The research reported in this publication was supported by the National Institutes of Health National Cancer Institute under award number R01CA187416. This work is supported in part by the U.S. National Science Foundation under the CyberTraining (OAC_1730250) and MRI (OAC_1726023) programs.
Handouts
Keywords
Protons, Image Processing, In Vivo Dosimetry
Taxonomy
TH- External Beam- Particle/high LET therapy: Range verification (in vivo/phantom): prompt gamma/PET
Contact Email