Correlating scored daily anatomical changes to in-vivo EPID dosimetry and cone beam CT based dose calculations

Abstract

At the Antoni van Leeuwenhoek Hospital/Dutch Cancer Institute (NKI-AvL) in Amsterdam, inter-fractional anatomical changes during the course of radiotherapy are monitored using cone beam CT scans, taken prior to irradiation. These scans are assessed visually, and the fractions are scored according to a 'traffic light protocol'. Based on the magnitude of change, a green, yellow, orange or red colour, in increasing order of severity, is assigned to the fraction.

The goal of this work was to ascertain if the colour of the traffic lights, which were assumed to be indicators of anatomical change, correlate to changes in dosimetry for H\&N VMAT treatments, as well as lung IMRT treatments.

The in-vivo EPID dose was reconstructed in the patient for each fraction, using a back-projection algorithm that is used clinically at the NKI. Calibrated CBCTs of each fraction were obtained using DIR or anti-scatter grid methods researched at the NKI, which were then imported to a TPS to obtain the fraction dose. These two modes of dosimetry were compared against each other, as well as against the traffic light colours for H\&N treatments. For lung treatments, due to unavailability of CBCT based dose data, only EPID dosimetry was used; two different models of the back-projection algorithm were compared in this case. γ index and DVH metrics were used to express deviation in the dose distributions.

Deviations over successive fractions for 18 H\&N treatments were studied. The traffic light protocol correlated poorly with CBCT based dose and EPID reconstructed dose (ρ = 0.33 and 0.35 respectively). The CBCT and EPID dose correlated with each other quite strongly (ρ = 0.72), however the EPID dose was more sensitive in its fluctuations.

Deviations for 98 IMRT lung fractions were studied. The traffic light protocol correlated even more poorly with the EPID reconstructed dose than in the H\&N study (ρ = 0.18). The calculated transmission model of the EPID was found to exaggerate the deviations in comparison to the measured transmission model. Since VMAT innately uses the calculated transmission model, this explains the sensitivity of the EPID results seen in the H\&N study.

We have shown that the traffic light protocol does not correlate with dosimetric changes, due to differences in assessment criteria. 15 out of 18 H\&N treatments showed moderate (ρ ≥ 0.4), if not strong, correlations between deviations of EPID reconstructed dose and CBCT based dose, strengthening the EPID's applicability for in-vivo dosimetry.