Abstract Book

S214

ESTRO 37

OC-0415 Real-time dose reconstruction for moving tumours in stereotactic liver radiotherapy S. Skouboe 1 , T. Ravkilde 2 , C.G. Muurholm 3 , E. Worm 2 , R. Hansen 2 , B. Weber 1,4 , M. Høyer 4,5 , P.R. Poulsen 1,5 1 Aarhus University Hospital, Department of Oncology, Aarhus, Denmark 2 Aarhus University Hospital, Department of Medical Physics, Aarhus, Denmark 3 Aarhus University, Department of Physics and Astronomy, Aarhus, Denmark 4 Aarhus University Hospital, Danish Center for Particle Therapy, Aarhus, Denmark 5 Aarhus University Hospital, Department of Clinical Medicine, Aarhus, Denmark Purpose or Objective The delivered dose in radiotherapy (RT) can differ substantially from the intended dose due to organ motion. Reconstruction of the actually delivered dose to a moving tumour could serve as an important tool for quality assurance and identification of cases where plan adaptation is needed. In this study, we developed and tested real-time reconstruction of the dose delivered to moving liver tumours during stereotactic body RT (SBRT). Material and Methods A software program, DoseTracker, was developed for real-time reconstruction of the dose delivered to any set of moving points within a body contour during RT using a modified pencil beam algorithm. In this study, DoseTracker was used offline for real-time reconstruction of the delivered tumour dose in respiratory-gated liver SBRT treatments guided by implanted Calypso transponders. Connection to the accelerator was simulated by sending a 21Hz data stream with the accelerator parameters (gantry, MLC positions, MU increment, etc.) and transponder positions of the actual treatment to DoseTracker. The dose was reconstructed in a calculation volume that included the CTV and was assumed to move rigidly with the transponders. Dose reconstruction was performed for four liver SBRT patients without motion (planned dose) and with the tumour motion of the actual gated three-fraction treatments and of simulated non-gated treatments. All reconstructed doses were imported into the treatment planning system (TPS, Eclipse) and compared with a previously published dose reconstruction method that emulates target motion as multiple isocenter shifts in the TPS. Since DoseTracker currently assumes water density in the patient, the TPS calculations were performed with both actual CT densities and water density. The motion-induced reduction in CTV D95 relative to the plan (ΔD95) was compared between DoseTracker and the TPS. Results Depending on tumour volume the real-time dose reconstructions were performed for 15k-93k calculation points with 2-5Hz mean frequency. Fig 1 shows reconstructed doses for the fraction with largest motion. Although DoseTracker underestimated the absolute dose level for this case ΔD95 (15.5 %-point) was in good agreement with the TPS water density calculation (14.1 %-point) and fair agreement with the TPS CT density calculation (12.2%) (Fig 2A). The root-mean-square error in the DoseTracker ΔD95 over all 24 motion-including dose reconstructions was 0.6 %-point when compared with TPS water density calculations and 1.2 %-point when compared with TPS CT density calculations (Fig 2B).

between, & postscan) are assigned based on the time the treatment beam was on (read from an in-house treatment log file) relative to the imaging time. Finally, ImStat exports the labels, setup data and imaging times to MS Excel for analysis via a VBA script. The script reads scan labels to, for example, evaluate IF motion and TT. We analyzed all SBRT lung cancer patients from 2009 to 2017 treated to a prescribed dose of 3×18Gy, 5×12Gy, or 8×7.5Gy. Image guidance involved a correction CBCT, a validation CBCT prior to treatment, an inline CBCT for each treatment arc, and a post RT CBCT, used only when intra-arc does not work. A treatment was automatically The tumor and spine IF motion and TT were evaluated based on the validation CBCT and the second inline CBCT or post RT CBCT. The average TT per patient was then correlated with the patient mean IF vector length by regression analysis. Results Data from 20829 scans of 1501 treatments (1271 pts) were automatically retrieved, labeled and exported to Excel within 2.25 hours using ImStat on a 64-bit 3.40GHz Intel® Xeon® PC. The VBA script took 5 mins to select the focus group to 16516 scans of 1149 treatments (1022 pts) and analyze the data (Table 1). Figure 1 shows the mean IF motion vector length vs TT ( p <0.001). The average TT was 12.2 min with a SD of 4.8 min. The wide TT range was due to a technique change from IMRT to VMAT that occurred within the observed period. excluded by the script from the selection when: a) the site name was not hypofractionated lung; b) the number of fractions was not 3, 5, or 8; c) at least one fraction had less than 3 CBCTs; d) there was no validation CBCT; e) ImStat failed to label at least one CBCT.

Conclusion ImStat is a powerful tool for fast automatic collection of image registration data, as demonstrated in this large 8.5-year dataset. For SBRT lung patients, the example analysis found a moderate positive correlation between tumor and bone intrafraction motion with treatment delivery time.