The Power of GPUs in Machine Learning to Improve Proton Beam Therapy for Cancer Treatment

Abstract

Proton beam therapy utilizes proton beams to treat cancerous tumors while avoiding unnecessary radiation exposure to surrounding healthy tissues. Real-time imaging of the proton beams while they travel through a patient’s body can make this form of radiotherapy more precise and safer for the patient. The use of a Compton camera is one proposed method for real-time imaging of the prompt gamma rays that are emitted by the proton beams. Unfortunately, some of the Compton camera data is flawed, and the image reconstruction algorithm yields noisy and insufficiently detailed images to evaluate the proton delivery for the patient. Machine learning can be a powerful tool to clean up the Compton camera images. Previous work used a deep residual fully connected neural network, but the use of recurrent neural networks (RNNs) has been proposed, since they use recurrence relationships to make potentially better predictions. In this work, RNN architectures using two different recurrent layers are tested, the LSTM and the GRU. Although the deep residual fully connected neural network achieves over 75% testing accuracy and our models achieve only over 73% testing accuracy, the simplicity of our RNN models containing only 6 hidden layers as opposed to 512 is a significant advantage.This will also cause the time to load the model from the disk to be significantly faster, potentially enabling the use of Compton camera image reconstruction in real-time during patient treatment. A graphics processing unit (GPU), known to perform complex math calculations to display high-quality graphics, could enable the use of this approach in a clinical setting since they are small in size and affordable.