Room: AAPM ePoster Library
Purpose: We propose a model for estimating dosimetric uncertainties as a consequence of discrepancies in the Deformable Image Registration (DIR) process when accumulating dose for adaptive Pencil Beam Scanned (PBS) proton therapy.
Methods: Three components are considered: (i) deformable vector fields (DVFs) derived from different DIR algorithms; (ii) a patient specific linear regression model of DVF uncertainty, and (iii) dose gradients from each fraction dose distribution. The model is constructed for the first fraction only, by correlating the magnitude of DVFs calculated with a reference algorithm (B-spline) to DVF uncertainty, quantified as the voxel-wise difference between the largest and smallest vector magnitudes of the different DVFs resulting from the different DIR algorithms. In subsequent fractions, only the reference (B-spline) DVF is applied, with dose uncertainties from its use being calculated by the multiplication of the model predicted DVF uncertainty with the dose gradients. For validation, adaptive PBS proton therapy plans were accumulated for seven lung cancer patients, for which nine repeated (fraction) CTs were available. As a ‘ground-truth’ (GT), each fraction specific dose distribution was warped back to the planning CT using the five different DIR algorithms, and the voxel-wise max-min dose uncertainty calculated. Voxel-wise max-min dose uncertainties were then calculated using the DVF uncertainty model and were compared to those from this GT scenario.
Results: Mean differences between predicted and GT uncertainties were 2.3±6.0% over all seven patients (each with nine CTs and six structures). For all non-zero dose inside the body, uncertainty could be predicted to 2.5±8.0% of the GT uncertainty.
Conclusion: We propose a method for estimating DIR induced uncertainties in dose accumulation for adapted PBS proton therapy, with predicted uncertainties being within 3% of the ground truth. The model will be further improved by including additional parameters such as DVF directional uncertainty and image feature information.
Funding Support, Disclosures, and Conflict of Interest: This project is funded by Krebsliga Schweiz.