Session Item

The purpose of pre-treatment VMAT QA is to ensure correct delivery of a treatment plan.  In our centre, where the TPS (Varian Eclipse) and R&V system (Elekta Mosaiq) do not share a database, these checks are especially important.

We intend to switch from Varian Portal Dosimetry to Sun Nuclear PerFraction for these checks.  These systems are conceptually similar; a predicted dose plane is compared to a measured dose plane derived from an EPID image.  We wish to assess the relative performance of the systems before switching.

In this work, the abilities of PerFraction and Portal Dosimetry to detect various types of deliberate VMAT delivery error were investigated.  The impact of such errors on patient dose was also assessed.

This is believed to be the first reported use of receiver operating characteristic (ROC) analysis to assess the performance of PerFraction.

This work used 10 treatment plans for various clinical sites, at 6 MV on a TrueBeam linac.  The predicted dose plane was calculated in both PerFraction and Portal Dosimetry for the unaltered plans. 

Modified versions of the plans were also created.  Changes were made to the total MU, central MLC positions, collimator angle and beam energy.

Unmodified treatments were delivered along with the modified versions.  EPID images were acquired during delivery and analysed using both systems.  Gamma analysis was used to compare the measured dose plane to the predicted dose plane for the unmodified plan.

To compare the error detection performance of the systems, ROC analysis was used.  The gamma pass rates for the modified and unmodified plans were used to construct ROC curves.  Greater area under the curve (AUC) indicates better error detection performance.

Modified plans were imported to Eclipse to assess the effect on patient DVHs.

In both systems, larger errors had higher detectability.  For machine output changes and beam energy changes, Portal Dosimetry had better error detection performance than PerFraction.  For MLC and collimator errors, the systems had comparable performance.  Table 1 gives AUC values for each system and error type.

All changes in patient DVH metrics for MLC shifts were found to be <2%.  Energy errors had a major impact on patient dose, up to around 20% for some metrics.  Collimator angle errors had an intermediate effect.

Table 1: AUC values for all ROC analyses.  The system with better performance is highlighted in green.

For some error types, the performance of PerFraction appears somewhat worse than Portal Dosimetry in the situations investigated.  For these cases the clinical impact is small, or we have other systems capable of detecting these errors.  We have therefore decided to implement PerFraction for routine pre-treatment QA as it performs adequately and gives independence from Varian systems and greater efficiency through automation of image retrieval and analysis.