(Wednesday, 8/1/2018) 10:15 AM - 12:15 PM
Room: Karl Dean Ballroom B1
Purpose: Advanced intra-fraction motion management strategies benefit from short-term predictive models to support real-time corrective actions. The goal of this research is to investigate the potential of Markov and semi-Markov models to predict the anatomical trajectory in real time using cine-MR images acquired during MR-guided radiation delivery. The intended application is real-time tracking and re-optimization of intensity-modulated delivery.
Methods: Organ motion of 9 stomach cancer patients was tracked retrospectively using local image features corresponding to the stomach boundary in 2D sagittal cine-mode MR, as well as global image features obtained via applying dimensionality reduction techniques to the region of interest on the MR frames. Complete respiratory cycles were modeled using an eight-parameter model that describes the motion amplitude and speed within the cycle. The set of cycles were clustered into distinct subgroups, and a Markov model was developed to describe the probabilistic transitions between those subgroups. Within each subgroup, a semi-Markov model was used to model the probabilistic transitions between different motion phases. Finally, a Bayesian model was used to continuously update the projected trajectory upon the arrival of each MR frame.
Results: The performance of the proposed predictive model with different motion traces is tested on the cine-MR data acquired for 9 stomach cancer cases using a cross-validation technique. The results are also compared against those obtained from the application of linear prediction models. Sensitivity analysis was performed on different model parameters. Results show that the proposed Markov model yields a higher prediction accuracy than the linear models.
Conclusion: In this research, a nested (semi-) Markov model was developed to predict the intra-fraction motion as seen in cine-MR images acquired during MR-guided radiotherapy. Results demonstrate that Markov processes are effective in predicting the short-term trajectory of intra-fraction motion.
Funding Support, Disclosures, and Conflict of Interest: National Science Foundation, Award Number 1662819
Keywords
Stochastic Processes, Organ Motion, Image-guided Therapy
Taxonomy
IM/TH- MRI in Radiation Therapy: MRI for treatment planning
Contact Email
