Christina Sarosiek ’16, Ph.D., is a postdoctoral fellow at the Medical College of Wisconsin (Milwaukee) researching deep learning methods applied to radiation oncology.
Effective radiation treatments for cancer rely heavily on frequent medical imaging to be able to localize the tumor volume and visualize the surrounding healthy tissues. Over the course of a 6–8 week treatment cycle, shifting anatomy, especially in the abdomen, can cause a once optimal treatment plan for a given day to be inaccurate for the next day. Ideally, a new image of the anatomy would be taken each day of treatment and the plan adapted to the anatomy on that day. However, current treatment planning workflows take several hours to complete, making daily replanning infeasible.
Dr. Sarosiek’s current work uses deep learning, a form of artificial intelligence, to automatically correct suboptimal auto-segmented contours as part of an end-to-end solution to automatically perform daily online adaptive replanning for MR-guided radiotherapy. With this work, the goal is to continue to make radiation therapy a safer and more effective cancer treatment.