Simulated real-time dose reconstruction for moving tumors in stereotactic liver radiotherapy

In radiotherapy, tumor motion may deteriorate the planned dose distribution. However, the dosimetric consequences of the motion are normally unknown for individual treatments. We here present a method for real-time motion-including tumor dose reconstruction and demonstrate its use for simulated stereotactic body radiotherapy (SBRT) of patients with liver cancer previously treated with Calypso-guided gating.

Real-time motion-including dose reconstruction was performed using in-house developed software, DoseTracker, on offline replays of previous clinical treatments. The patient cohort consisted of fifteen patients previously treated in our clinic with three-fraction SBRT to the liver using conformal or IMRT plans. The tumor motion at treatment was monitored with implanted electromagnetic transponders. The dose reconstruction was performed for both the actual gated treatments and simulated nongated treatments using a 21 Hz data stream containing accelerator parameters and the recorded motion. The dose was reconstructed in the same calculation points within the planning target volume (PTV) as used by the treatment planning system (TPS). The reconstructed doses were compared with calculations performed in the TPS, in which the motion was modeled as a series of isocenter shifts. The comparison included point doses as a function of treatment time and the dose volume histogram (DVH) for the clinical target volume (CTV). The motion-induced reduction in the dose to 95% of the CTV, $$ , and in the mean CTV dose, ΔD_Mean , was compared between DoseTracker and the TPS for each simulated fraction. DoseTracker currently assumes water density within the patient contour, so for comparison, the TPS calculations were performed with both CT density and water density. The calculation times were additionally analyzed.

Dose reconstruction was carried out for ninety SBRT sessions with calculation volumes ranging from 9.9 to 366.4 cm³ and median calculation times of 55-155 ms (equivalent to 18.2-6.5 Hz). Time-resolved trends of doses to a single calculation point in the patient were well replicated and dose differences between actual and planned calculations matched well. ΔD_Mean had a range of -0.1%-30.7%-points and was estimated by DoseTracker with a root-mean-square deviation (RMSD) to the TPS calculations of 0.43%-points (water density) and 0.79%-points (CT density). Similarly, $$ had a range of 0.0%-35.2%-points and was estimated by DoseTracker with an RMSD of 0.80%-points (water density) and 1.33%-points (CT density). DoseTracker predicted losses in tumor dose coverage above 5%-points with high sensitivity (91.7%) and specificity (97.6%).

Real-time dose reconstruction to moving tumors was demonstrated on offline replays of previous clinical treatments. DVHs of actually delivered dose are made available immediately after the end of treatment fractions. It shows promising results for liver SBRT with accurate estimation of CTV dose deteriorations caused by motion during treatment.