Session Item

Dose deterioration caused by target motion during radiotherapy can be a major concern for proton pencil beam scanning (PBS). Real-time motion-including dose reconstruction during treatments would enable detection of dose deviations and allow informed decisions and actions during treatment. Here, we create and experimentally validate a real-time proton dose reconstruction algorithm that accounts for dynamic motion.

Previously in-house developed software capable of online real-time motion-including dose reconstruction was expanded from photon to proton therapy. The software continuously calculates the dose to a moving target based on live-streamed treatment machine parameters and target positions. The proton dose reconstruction uses a pencil beam algorithm with density variations modelled by water equivalent path lengths calculated by real-time voxel traversal ray tracing through a CT scan. The pencil beam kernels for the algorithm were established by Monte Carlo (TOPAS) simulation of single spots in water in 2 MeV beam energy steps. Each kernel was parameterized as a depth dose curve and a depth dependent 2D Gaussian beam profile.

For experimental validation of the real-time dose reconstructions, a proton PBS plan was delivered to an ionization chamber array (MatriXX, IBA) placed on a motion stage (Quasar, Modus QA) with 5 cm solid water for build-up (Fig 1). The plan had a single energy layer (140 MeV) with 225 spots spaced 4mm apart in a square grid. Measurements were made twice with and once without applying a 40mm peak-to-peak sinusoidal 1-D motion with a 5s period. The ion chamber array measured the 2D dose distribution in dose frames with 7.62 mm spatial resolution and 10Hz temporal resolution.

After the experiments, the motion stage position was determined for each spot delivery in the motion experiments by comparison of the dose frames with corresponding dose frames from the static experiments. This was combined with machine log files to generate a data stream that each contained the position, energy and MU of each spot and the position of the motion stage. This data stream was broadcasted to the dose reconstruction software, which calculated the dose offline, but in real-time, both with and without motion. Real-time ray tracing was done through a synthetic CT matrix from the experimental setup geometry.

Measured and reconstructed doses were compared by 2%/2mm gamma failure rate evaluations with the measured doses as reference.

The absolute dose matched well for the single layer plan with and without motion (Fig 2). The gamma failure rate was 0.0 % (static), 1.1 % (motion 1) and 2.3 % (motion 2). 98.5 % of the spots were calculated within 3.0 ms.

A method to perform real-time proton dose reconstruction for dynamic motion was developed and experimentally validated. It reproduced measured doses well offline. The dose reconstructions can be done online when live streaming of spot delivery and target motion becomes available.