Traditionally, medical images of a patient scheduled for radiotherapy are acquired only once during the treatment preparation phase, one or more weeks before the start of treatment delivery. These images are subsequently used for target definition and treatment plan optimization but only represent a snapshot of the patient's anatomy. Over the course of radiation therapy patients undergo continuous changes in posture, anatomy and biology due to both physiology and treatment response. Consequently, the actually delivered dose typically deviates from the optimal dose. Therefore, we focus on strategies to monitor these changes through an image feedback loop and adapt the treatment to optimize the treatment outcome in the presence of anatomical and functional changes.
In the past, we have co-developed cone-beam CT (CBCT) scanners integrated with the treatment machine to capture the patient's anatomy just prior to irradiation. Currently, we are investigating both hardware and software solutions to improve the CBCT image quality to more accurately capture anatomical changes. Similarly, treatment machine integrated magnetic resonance imaging provides superior soft tissue contrast and real time imaging. We develop strategies to accelerate MR acquisitions and respiratory correlated imaging strategies.
To quantify such changes over the course of therapy, (deformable) image registration algorithms are optimized facilitating both couch corrections to align the target to the planned position as well as contour propagation and dose accumulation to adapt the treatment plan. Similarly, repeat functional imaging is utilized to monitor and model radiation response of target and organs at risk.
Artificial intelligence plays in increasingly important role in many aspects of adaptive radiotherapy moving away from hand crafted features and regularization terms and replacing them with data driven algorithms.