Individualised dose mapping uncertainty estimation in the reirradiation setting

Deformable image registration (DIR) allows assessing radiation doses of previous treatment courses in reirradiation planning scans. However, DIR can introduce uncertainties in dose mapping. Estimating these uncertainties for individual patients remains a key challenge. We developed an individualised approach to estimate uncertainties in dose mapping using a commercially available treatment planning system.

For 54 patients who underwent reirradiation (27 head and neck, H&N, and 27 lung cases), we performed 16 DIRs per patient. We assessed each DIR geometrically using the mean distance to agreement (mDTA) between mapped contours and the corresponding structures of 30 to 34 organs at risk (OARs). We then estimated dose mapping uncertainties for structures where DIR resulted in plausible registrations, defined as mDTA ≤ 0.3 cm. We calculated the uncertainty of dose-volume-histogram estimates (mean and D0.1 cm³) and voxel-wise uncertainties using standard deviation (SD). We also tested the impact of using fewer registrations.

DIR resulted in plausible mappings for both patient cohorts' OARs, ranging between 78.6 % and 88.2 %, varying between patients and OARs. Mean dose uncertainties ranged between 0 Gy and 0.43 Gy, while larger differences were observed for D0.1 cm3, between 0 Gy and 0.78 Gy. As expected, large voxel-wise uncertainties were in regions of steep dose gradients. Performance between registration sets was similar.

We developed a method to estimate dose mapping uncertainties with integrated quality control for reirradiation. Differences between patients were observed, justifying the need for individualised DIR assessment. Few registrations were a pragmatic approach.