Model based control of magnetic resonance guided thermal treatments: Theory and experimental evaluation

A magnetic resonance-guided control system is a significant enabling tool in realizing the clinical implementation of thermal therapies such as in cancer treatment and thermally activated drug delivery. Distributed parameter and nonlinear dynamics, safety and actuation constraints, and dynamic parameter variations are the major challenges in the development of such a control system. Recent progress in noninvasive magnetic resonance (MR) thermometry has allowed us to use real-time, distributed MR temperature information for the development of an improved thermal therapy control system. In particular, an image-based feedback control approach with dual objectives of satisfying clinical efficacy and safety requirements has been developed and evaluated using simulations, and in-vitro and in-vivo experiments. The efficacy requirement is formulated in the form of the control of thermal dose---a measure characterized by an integral of a nonlinear function of target temperature evolution. The normal tissue safety objective is incorporated in the controller design by imposing constraints on temperature elevations at clinician-selected normal tissue locations. In order to realize a broad applicability of the developed controller for a variety of heating modalities, control systems for both stationary and moving power deposition actuation fields have been developed. The stationary transducer control system has a cascade structure with the main nonlinear dose controller dynamically generating the reference temperature trajectory for the secondary, constrained, model predictive temperature controller. It is shown that the developed control approach allows near minimum-time delivery of the prescribed thermal dose to the target, which is a novel and clinically desirable feature. Controller results from simulations and experiments with tissue phantoms and in-vivo canines with ultrasound actuation and MR thermometry feedback are presented. The results demonstrate that a robust, near time-optimal control of the thermal dose is achievable without violating normal tissue temperature constraints despite a substantial plant-model mismatch. To extend the applicability of the stationary transducer controller, a mobile actuation controller for the treatment of geometrically complex and large targets is developed and validated using simulations. The moving transducer treatment approach uses a sliding-mode controller with a sliding manifold designed to meet the specified treatment efficacy and safety objectives. Simulation results show that the controller-generated transducer trajectory and power result in simultaneous activation of multiple normal tissue constraints while delivering the prescribed tumor thermal dose.