Session Item

Proton therapy with pencil beam scanning (PBS) in the thorax and abdomen is vulnerable to dose distortions caused by respiratory motion. The motion effects may be reduced by rescanning or treatment in breath-hold. Breath-sampling rescanning, where the rescanning of each layer is designed to cover the entire breathing cycle, provides the most efficient interplay effect mitigation, but it relies on accurate predictions of the delivery time of each PBS spot. Such predictions are also useful when evaluating the suitability of PBS plans for breath-hold, and they will be crucial for plan optimization in future proton FLASH treatment planning. Here, we develop a simulator for accurate prediction of proton PBS delivery times.

A software tool for simulation of proton PBS fields was created and tested at a proton treatment gantry (ProBeam, Varian). The simulator modeled the timing of proton spots as a waiting time that depended on energy and distance from the preceding spot plus a delivery time equal to the number of monitor units (MU) divided by an energy layer-specific MU-rate. The MU-rate was estimated from the cyclotron beam current, which was simulated by taking interdependencies of all energy layers, cyclotron current restrictions and the minimum allowed spot duration (3 ms) into account. The parameters for the simulator were first fitted to delivery log files from a training dataset with 18 PBS plans that covered the clinical energy range (80-220 MeV in 3MeV steps) and spanned wide ranges of spot MUs (1-80 MU) and spot distances (5-180 mm). The spots in the training plans were delivered in a spiral pattern with increasing spot distances (Figure 1.A). Next, the simulator was tested by predicting the delivery time for each individual spot in 83 clinical treatment fields with a total of 1694 energy layers. The simulated delivery time of each layer was compared with the actual delivery times as given by the log files.

Figures 1.B and 1.C show the waiting time of spots as function of energy and distance from the preceding spot logged during delivery of a training dataset plan. The cyclotron current in an energy layer varied from 74nA to 550 nA and was predicted with a root-mean-square (RMS) error of 1.6% (Figure 2.A). The duration of individual layers varied from 0.003s to 3.0s. Figures 2.B-2.D compares the simulated layer durations with the logged durations. The layer duration was predicted with RMS errors of 0.081 s (total layer duration), 0.003 s (total time of spot shifts in the layer) and 0.080 s (total beam-on time in the layer).

A simulator of proton PBS delivery was developed and shown to predict the timing of PBS delivery accurately. It has useful applications in the clinic where the lower MU limit defines the treatment time and in the design of rescanning techniques and evaluation of proton plans for breath-hold treatments. It could also play an important role in optimization of FLASH proton plans, where accurate dose rate predictions are crucial.