Radiation therapy is an effective treatment for a wide range of cancers but delivering the precise and customized dose of radiation while minimizing exposure to normal and healthy tissues can be challenging. One of the major challenges is the respiratory motion. Traditional MRI, however, can be time-consuming. Long acquisition times can be inconvenient for patients and limit the number of patients that can be evaluated in a day. Additionally, long acquisition times can also limit the ability to perform adaptive tasks, such as adaptive radiotherapy by adjusting the radiation beams in real-time, based on the motion of the tumor or organs-at-risk. To address these challenges, this project is using accelerated MRI and deep learning-based MRI reconstruction methods. Accelerated MRI involves sampling less measurements, known as undersampling, using various patterns. Deep learning-based MRI reconstruction methods reconstruct high-quality images from undersampled MRI data allowing for faster imaging times and improved image quality. The goal of this project is to enhance the precision of radiotherapy by using the MR-linac machine and MRI reconstruction methods with deep learning. This will improve the accuracy of radiation therapy by reducing the risk of motion-induced artifacts, and ultimately improving the outcome for patients.

This project has three main steps:

1. Learning an optimal deep learning-based accelerated MRI reconstruction algorithm.
2. Learning optimal real-time active acquisition schemes using a deep learning algorithm to optimize the acquisition process
3. Tracking tumors and organs-at-risk directly from the measurements, and link their motion to changes in the radiation beams. By doing this, the project can improve the accuracy and efficiency of the radiation delivery, and reduce the risk of normal tissue damage.​