Promising Hyperparameter Configurations for Deep Fully Connected Neural Networks to Improve Image Reconstruction in Proton Radiotherapy
Published in IEEE International Conference on Big Data, 2021
We conduct hyperparameter grid search for deep fully connected neural networks with residual blocks to classify prompt gamma interactions in Compton camera data. The optimized compact networks achieve competitive accuracy for real-time proton beam imaging in cancer treatment.
Recommended citation: S.A. York, A.M. Ali, D.C. Lashbrooke Jr., R. Yepez-Lopez, C.A. Barajas, M.K. Gobbert, J.C. Polf. "Promising Hyperparameter Configurations for Deep Fully Connected Neural Networks to Improve Image Reconstruction in Proton Radiotherapy." IEEE Big Data 2021.
Download Paper