Deep Learning in the Context of Radiotherapy/Medical Physics (Landry)

Project description

In ion therapy of cancer, beams of protons or heavier ions are used to treat tumors while sparing healthy tissue by exploiting their finite range. Large projects in Germany are attempting to combine ion beams with magnetic resonance image guidance to better distinguish tumors and tissues during irradiation. This brings two challenges: i) dose calculation on MR images and ii) proton transport in the presence of magnetic fields to account for the curved trajectories due to the Lorentz force. Classically i) requires the conversion of MR images into CT images and ii) the use of lengthy Monte Carlo dose calculations. The goal of this project is to bypass both i) and ii) by using deep learning to calculate proton dose distributions directly on MR images by exploiting a large library of pre-calculated Monte Carlo dose distributions in patient anatomies. Deep learning methods designed for sequences, such as long short term memory (LSTM) networks or transformers will be used to convert a series of 2D MR slices in beam’s eye view into proton pencil beam dose distributions. The project offers the possibility to acquire advanced deep learning-based image processing skills as well as experience with Monte Carlo simulation.