Program Information
Robust Model Predictive Control for Adaptive Radiation Therapy
M Boeck*, A Forsgren , K Eriksson , J Karlsson , KTH Royal Institute of Technology, Stockholm, Stockholm
Presentations
TU-L-GePD-J(A)-5 (Tuesday, August 1, 2017) 1:15 PM - 1:45 PM Room: Joint Imaging-Therapy ePoster Lounge - A
Purpose: To set up a framework integrating optimal control and robust model predictive control (MPC) into adaptive radiation therapy (ART) planning in order to mitigate the impact of interfractional geometric variations on the dose distribution. The proposed discrete-time stochastic formalism is designed to dynamically adapt to spatiotemporal changes occurring throughout a patient’s treatment.
Methods: The authors present a robust MPC framework for ART based on scenario-based stochastic optimization for a series of simulated treatments on an idealized patient phantom. While robust MPC is widely used in process control, its look-ahead feature is hardly exploited in ART planning. In this discrete-time control framework, the received dose in the target and two OARs are used to model the system states, while the fluence intensities are represented by the controls. The geometric variations which disturb the system are discretized into scenarios and modeled as independent and identically distributed random variables. In contrast to the current approach to adaptive planning, this framework utilizes the measured dose at each fraction as state feedback in order to determine the optimal control policy for the next fraction by minimizing the expected cost function for different prediction horizons.
Results: Monte Carlo simulations for different prediction horizons are preformed to test all potential uncertainties, which indicate that sufficient target coverage is maintained throughout the treatment, and that the state trajectory predictions are quite accurate. Our preliminary computational results show that in comparison to the simulations with the optimal control algorithm and the conventional non adaptive approach, MPC improves the probability to achieve target coverage without compromising the OARs in the presence of random uncertainties.
Conclusion: The initial computational study indicates that combining MPC with ART dynamically adapts to uncertain interfractional variations and is able to estimate the future motion pattern in order to maintain target coverage and OAR protection.
Contact Email: