Comprehensive fluence delivery optimization with multileaf collimation

Abstract

We propose a novel comprehensive model of the dynamic multileaf collimator (MLC) sequencing problem under the sliding window technique. While the majority of research regarding MLC sequencing has lain dormant for years, leaf sequencing is currently done via ‘black-box’ implementations in clinical cancer treatment planning software. As such, it is unclear which leaf motion and fluence transmission parameters are included in these models, given their widely varying analytic and heuristic treatment in the existing literature. We hypothesize that an explicit, comprehensive model may fill an essential role in further research into intensity modulated radiation therapy and volumetric modulated arc therapy. To this end, we consolidate considerations of leaf motion (maximum velocity, finite acceleration), transmission (through-leaf, inter-leaf via tongue and groove) and novel formulations for penumbra across both dimensions of the field. In addition, we formulate our model to utilize these varying transmission effects to optimally sequence leaves with the goal of creating a fluence with pixel size smaller than the narrowest leaf width. To evaluate the proposed model, we have optimized MLC leaf sequencing on 25 prostate, 25 head and neck, 25 pelvis and 35 breast cancer fluence fields. The output sequenced fluences with and without constraints were compared with the corresponding reference fluences, respectively, by the performance of root mean square error and gamma index analysis. The acceptance criteria of 0.5%/0.5 mm and 1%/1 mm were used with a 0%, 5% and 10% low intensity threshold, respectively. Under the consideration of aforementioned constraints, the model can sequence the reference fluence successfully with the percentage of gamma passing rate ranging from 82.23 ± 3.89 to 99.78 ± 0.16 at 0.5%/0.5 mm and from 88.24 ± 1.89 to 99.98 ± 0.03 at 1%/1 mm across all low intensity thresholds and four treatment sites.