Click here to


Are you sure ?

Yes, do it No, cancel

Development of a Centralized Motion Prediction Model for Motion Management in Radiation Therapy

J Kim*, A Tai , X Li , H Zhong , Medical College of Wisconsin, Milwaukee, WI


(Tuesday, 7/16/2019) 10:30 AM - 11:00 AM

Room: Exhibit Hall | Forum 5

Purpose: A strategy to manage intrafraction motion during radiation therapy is to track the motion of an anatomic reference landmark (ARL). This work aims to develop a centralized model to predict motion of anatomical structures based on the tracked motion of the ARL.

Methods: The model was developed based on 10-phase 4DCT datasets. Each 4DCT was used to generate a mid-position (MidP) image, representing the time-weighted, centralized anatomy position during the 4DCT-breathing cycle. A deformable image registration (DIR) was applied to register the MidP-CT with each of the 10-phase images. The motion state Š of any anatomical structure can be modeled with Š = S + α·Dφ, where S is reference state (MidP), α is amplitude, and Dφ represents the output obtained by interpolating the input DVFs (MidP-CT to each 4DCT-frame) in correspondence of phase φ. Two typical anatomical landmarks nearby chest-wall and diaphragm, respectively, were selected as the ARL. By tracking the ARL, the respiratory parameters (α and φ) were updated to build a centralized, real-time prediction model. The consistency of the models developed with the two ARLs was evaluated using 100 feature points.

Results: The results showed that the target registration error of the centralized model was 1.9±1.1, 2.0±0.9, and 1.6±0.8 mm respectively for 3 lung cancer patients. The mean difference of the models developed with the chest-wall and diaphragm ARLs, related to their deformation amplitudes, were 8.7%, 5.6% and 8.5% respectively, and their differences at the 100 evaluation points were 1.93±0.86, 1.94±0.95, 1.52±0.69 mm on average for the 3 patients.

Conclusion: The newly-developed real-time centralized model can predict target motion within 2 mm. The difference between two prediction models is limited. This technique could help ensure consistent target tracking when different 4D-images (4DCT or 4DMRI) are used during radiotherapy.


Organ Motion, Registration, Image-guided Therapy



Contact Email