Purpose The unpredictable interplay between dynamic proton therapy delivery and target motion in the thorax can lead to severe dose distortions. A fraction‐wise four‐dimensional (4D) dose reconstruction workflow allows for the assessment of the applied dose after patient treatment while considering the actual beam delivery sequence extracted from machine log files, the recorded breathing pattern and the geometric information from a 4D computed tomography scan (4DCT). Such an algorithm capable of accounting for amplitude‐sorted 4DCTs was implemented and its accuracy as well as its sensitivity to input parameter variations was experimentally evaluated. Methods An anthropomorphic thorax phantom with a movable insert containing a target surrogate and a radiochromic film was irradiated with a monoenergetic field for various 1D target motion forms (sin, sin4) and peak‐to‐peak amplitudes (5/10/15/20/30 mm). The measured characteristic film dose distributions were compared to the respective sections in the 4D reconstructed doses using a 2D γ‐analysis (3 mm, 3%); γ‐pass rates were derived for different dose grid resolutions (1 mm/3 mm) and deformable image registrations (DIR, automatic/manual) applied during the 4D dose reconstruction process. In an additional analysis, the sensitivity of reconstructed dose distributions against potential asynchronous timing of the motion and machine log files was investigated for both a monoenergetic field and more realistic 4D robustly optimized fields by artificially introduced offsets of ±1/5/25/50/250 ms. The resulting dose distributions with asynchronized log files were compared to those with synchronized log files by means of a 3D γ‐analysis (1 mm, 1%) and the evaluation of absolute dose differences. Results The induced characteristic interplay patterns on the films were well reproduced by the 4D dose reconstruction with 2D γ‐pass rates ≥95% for almost all cases with motion magnitudes ≤15 mm. In general, the 2D γ‐pass rates showed a significant decrease for larger motion amplitudes and increase when using a finer dose grid resolution but were not affected by the choice of motion form (sin, sin4). There was also a trend, though not statistically significant, toward the manually defined DIR for better quality of the reconstructed dose distributions in the area imaged by the film. The 4D dose reconstruction results for the monoenergetic as well as the 4D robustly optimized fields were robust against small asynchronies between motion and machine log files of up to 5 ms, which is in the order of potential network latencies. Conclusions We have implemented a 4D log file‐based proton dose reconstruction that accounts for amplitude‐sorted 4DCTs. Its accuracy was proven to be clinically acceptable for target motion magnitudes of up to 15 mm. Particular attention should be paid to the synchronization of the log file generating systems as the reconstructed dose distribution may vary with log file asynchronies larger than those caused by realistic network delays.

No abstract