Predictive Imaging Methodology for Real-Time MRI-guided Radiotherapy

Motivation: MRI-radiotherapy hybrid units can allow for the therapy beam to be guided to match the position and shape of moving anatomic structures. However, there is substantial equipment lag to control the beam position (100s of milliseconds). To target accurately, the future position and shape of the target must be predicted. Goal(s): The goal of this work is to predict real-time images 1 or 2 frames beyond the current image. Approach: PCA analysis can allow for images to be represented with a small number of coefficients. These can be extrapolated to a future time. Results: Predicted images were sufficient for radiotherapy guidance. Impact: Prediction of images beyond the current real-time image will allow a simple and accurate means of controlling beam position, overcoming issues with equipment lag. Because this work does not rely on prior characterizations, it is potentially robust against non-cyclic motions.