A mimetic algorithm for simultaneous multileaf collimator aperture shape and dosimetric optimization in CyberKnife robotic radiosurgery

A mimetic algorithm for the optimization of multileaf collimator aperture shape and beam weighting is proposed for the CyberKnife M6 delivery platform (Accuray Inc., Sunnyvale, CA). The problem of optimizing a patient-specific treatment plan necessarily requires the determination of the beam geometry, beam-defining aperture shape, and beam weighting. The robotic arm of the CyberKnife moves the linear accelerator through approximately half the solid angle, the multileaf collimator in the head of the accelerator provides for conformal aperture shaping of the radiation, and the linear accelerator generates megavoltage x-rays to irradiate the target. The multileaf collimator leaf positions are optimized, as well as beam weights, by an algorithm consisting of an EA combined with a CMA-ES intense local search. The generation of apertures uses a rule-based approach, based upon dosimetric feedback regarding the radiation dose distribution acquired during each iteration. The purpose of the optimization is to find a Pareto-optimal solution which takes into account the clinical decision-making calculus of the radiation oncologist. The algorithm was tested on CT data set for a patient with carcinoma of the prostate. Twenty algorithm runs were performed; the mean fitness value was 2.48 × 10⁵, and the standard deviation was 9.40 × 10⁴. In all instances, the algorithm produced clinically acceptable treatment plans, meeting the user-specified dose-volume objectives for the target and critical structures. Note that there is no commercial algorithm currently available for generating these types of treatment plans, so the present work demonstrates feasibility of a novel treatment planning approach.