About this Event
27 Graves Place, Holland, MI 49423-3617
A Deep learning-based Automatic Contour Refinement Method for MRI-guided Online Adaptive Radiotherapy
Radiation therapy is a cancer treatment that uses radiation to kill cancer cells and reduce tumor size. The radiation treatment plans rely on computed tomographic (CT) images of the patient taken prior to the first treatment day to accurately predict the dose delivered to the patient. However, daily changes in anatomy such as weight loss/gain or tumor shrinkage can reduce the accuracy of these treatment plans and lead to underdosing the tumor or overdosing healthy tissues. MRI guidance in radiotherapy allows for more frequent imaging without excess dose to the patient and has an additional benefit of improved soft tissue visualization in the abdomen. Online adaptive radiotherapy (OART) uses recent advances in artificial intelligence to quickly adapt the original treatment plan to the anatomy of the day. One major piece of the OART process involves creating new anatomical contours on the daily MR image. Deep learning-based automatic segmentation methods do not produce quality contours sufficient for treatment and require manual correction. Our group is developing an automated contour refinement workflow to improve contours after initial automatic segmentation with the overall goal to make MRI-guided radiotherapy treatments safer, more efficient, and more effective.
(Left) Example slice from an abdominal MRI taken with an MR-linac at MCW. (Right) Same MR slice cropped in to show the stomach. The blue contour was drawn manually by a physicist for treatment planning. In OART, the auto-segmentation method produces in correct contours, such as the one shown in green. We can use deep learning to improve the accuracy of the incorrect contours. The red corrected contour shows good agreement with that of the manually drawn contour in blue and thus can be used for treatment planning.
Christina Sarosiek (Hope College ’16) is a post-doctoral fellow at the Medical College of Wisconsin. She received her PhD in Physics in 2021 from Northern Illinois University where she studied the clinical applications of proton computed tomography and proton radiography. She is now applying deep learning to online adaptive radiotherapy by developing techniques to automatically correct anatomical contours.
Meet Dr. Sarosiek in this short interview-^
0 people are interested in this event