Interobserver variability of patient positioning using four different CT datasets for image registration in lung stereotactic body radiotherapy

Purpose To assess the impact of different reference CT datasets on manual image registration with free-breathing three-dimensional (3D) cone beam CTs (FB-CBCT) for patient positioning by several observers. Methods For 48 patients with lung lesions, manual image registration with FB-CBCTs was performed by four observers. A slow planning CT (PCT), average intensity projection (AIP), maximum intensity projection (MIP), and midventilation CT (MidV) were used as reference images. Couch shift differences between the four reference CT datasets for each observer as well as shift differences between the observers for the same reference CT dataset were determined. Statistical analyses were performed and correlations between the registration differences and the 3D tumor motion and the CBCT score were calculated. Results The mean 3D shift difference between different reference CT datasets was the smallest for AIPvsMIP (range 1.1-2.2 mm) and the largest for MidVvsPCT (2.8-3.5 mm) with differences > 10 mm. The Delta 3D shifts showed partially significant correlations to Delta 3D tumor motion and CBCT score. The interobserver comparison for the same reference CTs resulted in the smallest Delta 3D mean differences and mean Delta 3D standard deviation for a dagger AIP (1.5 +/- 0.7 mm, 0.7 +/- 0.4 mm). The maximal 3D shift difference between observers was 10.4 mm (Delta MidV). Both 3D tumor motion and mean CBCT score correlated with the shift differences (R-s = 0.336-0.740). Conclusion The applied reference CT dataset impacts image registration and causes interobserver variabilities. The 3D tumor motion and CBCT quality affect shift differences. The smallest differences were found for AIP which might be the most appropriate CT dataset for image registration with FB-CBCT.