(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: ion chambers cannot provide satisfactory proton radiography due to the limitation of proton scatter. This work proposes a novel proton radiography method using deep learning and the newly designed 4D array ion chamber detector to characterize patient anatomy and validate dose calculation algorithm.
Methods: detetor has 2 mm and 0.3 ms spatial-temporal resolutions of 128Lx128Wx64D channels. Deep convolutional neural networks (CNNs) were used to assist data analysis from the substantial amount of spatial-temporal measured data using 41x41 proton spots per 4 mm spacing. The anthropomorphic phantom was irradiated with an anterior beam and the exit dose was measured with the 4D detector and conventional MatriXXPT. Water equivalent thickness (WET) map were derived by matching measurements with RayStation 8A dose calculation.
Results: WET maps agree between 2 mm for the most parts and large discrepancies of 4-8 mm often happen near the edge of dens bone that involve high heterogeneity. The results indicate that the proton radiography image can be improved with 4D detector by de-convolution of proton scatter involved in MatriXX PT measurements. The difference of WET maps increased with the complexity of clinical sites.
Conclusion: newly designed 4D detector featured in spatial-temporal measurement capability can utilize the pencil beam scanning feature of proton beam, improve accuracy of derived WET maps, shorten the acquisition time and reduce the radiation dose.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
