MO‐C‐17A‐01: BEST IN PHYSICS (JOINT IMAGING‐THERAPY) – A Method to Estimate Deformable Dose Accumulation Errors in Patients with Cancers of the Head and Neck

Abstract

Purpose:The dosimetric uncertainties of deformable image registration (DIR) dose accumulation are not well understood. A clinically relevant method to estimate the dosimetric error of dose accumulation for head and neck (H&N) cancer patients is presented.Methods:Ten virtual H&N phantoms with known deformations were registered with a commercial DIR algorithm to determine the spatial errors that might be observed when performing DIR. The spatial error distributions obtained from the phantoms were then applied to 10 clinical H&N patients to simulate the potential errors that could occur. These errors were simulated by selecting ROI‐specific error vectors at random from the phantom error distributions and adding them to the deformation vector fields (DVFs) of the patient registrations. This method of error simulation was evaluated for its ability to accurately represent real DVF errors. Finally, dose was deformed using the DVFs of the patient registrations without any added errors (nonperturbed DVFs) and compared to the dose deformed using the simulated error DVFs.Results:The mean error introduced by the error simulation method was less than 1% for all evaluated DVH endpoints. For the patient cases, the following ranges were observed for the median simulated DIR dosimetric errors: brainstem D2%: 0% to 2%, cord D2%: 0% to 0.5%, mandible D2%: 0% to 0.5%, left parotid Dmean: 0.5% to 4.2%, right parotid Dmean: 0.2% to 1.3%. Maximum dosimetric errors were: brainstem D2%: 6.5%, cord D2%: 1.3%, mandible D2%: 1.1%, left parotid Dmean: 13%, right parotid Dmean: 35%.Conclusions:A method for simulating DIR errors for dose accumulation in clinical cases was presented and evaluated. The sample DIR algorithm assessed with this method showed that large dosimetric errors are possible when dose accumulation is performed and that DIR should be used with caution.This work was partially supported by a research grant from Accuray Inc.