[Paper Review] Fast approximate delivery of fluence maps: the single map case

David Craft, Marleen Balvert|arXiv (Cornell University)|Aug 9, 2016

Advanced Radiotherapy Techniques5 references3 citations

TL;DR

This paper proposes a continuous, non-convex optimization framework that jointly determines optimal leaf trajectories and dynamic dose rates for fast, accurate delivery of a single fluence map in IMRT using a multi-leaf collimator. By modeling the problem as a constrained optimization and solving it via an interior-point method with random initialization, the approach achieves near-optimal fluence map approximation within a fixed delivery time, with results showing significant improvement over standard methods, especially under time constraints.

ABSTRACT

In this first paper of a two-paper series, we present a method for optimizing the dynamic delivery of fluence maps in radiation therapy. For a given fluence map and a given delivery time, the optimization of the leaf trajectories of a multi-leaf collimator to approximately form the given fluence map is a non-convex optimization problem. Its general solution has not been addressed in the literature, despite the fact that dynamic delivery of fluence maps has long been a common approach to intensity modulated radiation therapy. We model the leaf trajectory and dose rate optimization as a non-convex continuous optimization problem and solve it by an interior point method from randomly initialized feasible starting solutions. We demonstrate the method on a fluence map from a prostate case and a larger fluence map from a head-and-neck case. While useful for static beam IMRT delivery, our main motivation for this work is the extension to the case of sequential fluence map delivery, i.e. the case of VMAT, which is the topic of the second paper.

Motivation & Objective

• To address the unsolved problem of optimizing dynamic leaf sequencing for a single fluence map under fixed delivery time and machine constraints.
• To develop a continuous, non-convex optimization model that jointly optimizes leaf trajectories and dose rate profiles for improved fluence map approximation.
• To provide a foundation for extending the method to sequential fluence map delivery in VMAT, where leaf position continuity between fractions is critical.
• To demonstrate the feasibility and effectiveness of the approach on clinical fluence maps from prostate and head-and-neck cases.
• To establish a framework that can be extended to include physical machine constraints such as leaf speed limits, dose rate limits, and leaf transmission.

Proposed method

• Formulates the dynamic fluence map delivery problem as a non-convex continuous optimization problem with constraints on leaf speed, dose rate, and leaf position.
• Uses an exposure function e(L,R,j) to map leaf positions (L, R) and bixel index j to the delivered fluence, incorporating partial and full blocking effects.
• Applies an interior-point method with random feasible starting points to solve the non-convex problem, leveraging gradient information for convergence.
• Imposes constraints on maximum leaf speed and dose rate, with the dose rate allowed to vary continuously over time.
• Models the coupling between leaf rows through time-varying dose rate, ensuring simultaneous optimization across all rows.
• Validates the method on a prostate and a head-and-neck fluence map, demonstrating robustness and convergence within hours.

Experimental results

Research questions

• RQ1 What is the optimal combination of leaf trajectories and dynamic dose rates that minimizes fluence map error for a single fluence map within a fixed delivery time?
• RQ2 How does the method perform under realistic linac constraints such as maximum leaf speed and dose rate limits?
• RQ3 Can the proposed optimization framework achieve near-optimal fluence map delivery when the problem is non-convex and coupled across rows via time-varying dose rate?
• RQ4 How does the method scale with increasing fluence map complexity, such as in large head-and-neck cases?
• RQ5 What is the impact of delivery time on solution quality, particularly when dose rate is not at its maximum?

Key findings

• The method achieves near-optimal fluence map approximation by jointly optimizing leaf trajectories and dynamic dose rates, with solution quality plateauing after a few hours of computation.
• For a 7-second delivery time, the optimal solution rarely exceeds 5 MU/sec, even though the maximum dose rate is 10 MU/sec, indicating that dose rate constraints are inactive in this regime.
• The approach significantly outperforms standard sliding window techniques, especially when delivery time is constrained, due to the ability to adapt dose rate and leaf motion dynamically.
• The method is robust across different fluence map types, successfully delivering both a prostate case and a larger head-and-neck case with high fidelity.
• The optimization framework is extensible to include physical machine effects such as leaf transmission, jaw motion, and minimum leaf gap constraints through modifications to the exposure function.
• The work lays a critical foundation for the second paper, which will address sequential fluence map delivery in VMAT by ensuring continuity of leaf positions between maps to avoid time-wasting repositioning.