Results: The number of the registrations that fulfilled the selection criteria varied largely with the patients and fractions, ranging from 5 to 109, emphasizing the need for individualized parameter settings for non-rigid registration. Over all patients, the average±1 SD of the 99th percentiles of the spatial uncertainties was 5.2±3.0 mm and was larger than 10 mm for three daily CT scans. The average variation in fractional dose (1 SD) over all voxels varied between 0.1 Gy and 1.1 Gy for the esophagus, 0.1 Gy and 0.6 Gy for the stomach and 0.01 Gy and 0.6 Gy for the duodenum. This translated into an uncertainty (1 SD) of the cumulative maximum doses of 0.04–0.6 Gy for the esophagus, 0.03–0.4 Gy for the stomach and 0.01–0.2 Gy for the duodenum. For one patient the cumulative maximum dose exceeded the clinical constraint (the planned dose did not)(Fig 1).

Conclusions: Even though all registrations led to visually acceptable results, the spatial and dose variation were considerable, which confirms the relevance of dealing with dose uncertainty caused by non-rigid registration. The results further emphasize the need of individualized parameter settings for non-rigid registration. This study is an important step to introduce non-rigid registration in a clinical setting, for instance for adaptive radiotherapy.