Optimization of spatiotemporally fractionated radiotherapy treatments with bounds on the achievable benefit

TL;DR

This paper introduces a novel optimization framework for spatiotemporal radiotherapy that computes bounds on achievable normal tissue dose reduction, demonstrating significant potential benefits in liver tumor treatments.

Contribution

It develops a new optimization model with bounds on normal tissue BED reduction, combining convex and nonconvex methods to evaluate treatment plans.

Findings

Spatiotemporal plans reduced mean liver BED by 12-35%.

Plans achieved 79-97% of the theoretical maximum BED reduction.

Local optimization plans are close to the global potential.

Abstract

Spatiotemporal fractionation schemes, that is, treatments delivering different dose distributions in different fractions, may lower treatment side effects without compromising tumor control. This is achieved by hypofractionating parts of the tumor while delivering approximately uniformly fractionated doses to the healthy tissue. Optimization of such treatments is based on biologically effective dose (BED), which leads to computationally challenging nonconvex optimization problems. Current optimization methods yield only locally optimal plans, and it has been unclear whether these are close to the global optimum. We present an optimization model to compute rigorous bounds on the normal tissue BED reduction achievable by such plans. The approach is demonstrated on liver tumors, where the primary goal is to reduce mean liver BED without compromising other treatment objectives. First a uniformly fractionated reference plan is computed using convex optimization. Then a nonconvex quadratically constrained quadratic programming model is solved to local optimality to compute a spatiotemporally fractionated plan that minimizes mean liver BED subject to the constraints that the plan is no worse than the reference plan with respect to all other planning goals. Finally, we derive a convex relaxation of the second model in the form of a semidefinite programming problem, which provides a lower bound on the lowest achievable mean liver BED. The method is presented on 5 cases with distinct geometries. The computed spatiotemporal plans achieve 12-35% mean liver BED reduction over the reference plans, which corresponds to 79-97% of the gap between the reference mean liver BEDs and our lower bounds. This indicates that spatiotemporal treatments can achieve substantial reduction in normal tissue BED, and that local optimization provides plans that are close to realizing the maximum potential benefit.