Fig 1.
Relationships of lipiodol position in 3D– and 4D–CBCT for all 105 fractions of all 18 patients in the three cardinal directions.
Solid lines are the linear fits. Corresponding Pearson’s correlation coefficients (r) are also indicated.
Fig 2.
Registration results are highly similar between 3D– and 4D–CBCT (displayed in same window / level).
Effect of motion artifact is more pronounced on the 3D–CBCT images (bottom) than on the time–weighted average 4D–CBCT images (top), as indicated by the green arrows. The lipiodol contrast, however, is visualized brighter on 3D–CBCT than on 4D–CBCT due to a tenfold increase of projection data.
Table 1.
Population statistics of the absolute / relative positions of the lipiodol with respect to the vertebra in 3D– and 4D–CBCT, and their difference of absolute position in terms of group means (GM), effective systematic (Σeff), and random (σeff) errors.
Fig 3.
Inverse cumulative probability plots of the absolute tumor prediction errors for all 105 fractions of all 18 patients (top), and for individual means (bottom).
Fig 4.
Relationships of lipiodol position between the lipiodol–diaphragm distance and the tumor prediction errors in three cardinal directions.
Solid lines are the linear fits with the corresponding Pearson’s correlation coefficients (r). The unfilled dot in craniocaudal direction (middle) indicated the outliner.
Fig 5.
Calculated margins for different treatment approaches.
For no image–guidance corrections and offline corrections (i.e. registration to bony anatomy) the margins were calculated using the estimated systematic (Σeff) and random (σeff) errors of absolute and relative lipiodol position from 3D– and 4D–CBCT (Table 1). Margins for the online 4D–CBCT using lipiodol registration were obtained assuming ideal soft–tissue tumor correction protocol with zeroed Σeff and σeff, while margins for the online lipiodol–based 3D–CBCT were obtained includingΣeff and σeff due to the offset between 3D– and 4D–CBCT.