Room: Karl Dean Ballroom A1
Purpose: To directly compare independent plan recalculation against measurement-based IMRT QA to see which performed better at detecting unacceptable plans.
Methods: Acceptability of IMRT delivery was assessed with 337 IROC head and neck phantoms previously irradiated as part of clinical trial credentialing, 18 of which failed to meet IROCâ€™s acceptability criteria. For each case, the institutionâ€™s IMRT QA report was abstracted to determine how well their clinical QA predicted the irradiation result (i.e., pass or fail). Each case was also independently recalculated by IROC using the institutionâ€™s DICOM data and Mobius 3D (with linac class-specific generic beam models) to determine how well the recalculation predicted the irradiation results. Comparisons between IMRT QA and recalculation were made using truth tables to determine sensitivity and specificity of each, including subdivision by IMRT QA device (EPID, ArcCheck, ion chamber, ion chamber + array, or MapCheck). ROC analysis was also performed to evaluate the accuracy of recalculation versus IMRT QA as the strictness of the criteria for flagging failures varied.
Results: For the 18 failing phantoms in the total cohort, measurement-based IMRT QA had a sensitivity of 6% (i.e., only 1 unacceptable plan was flagged). The recalculation approach had a sensitivity of 72% (flagging 13 unacceptable plans). When the results were subdivided by IMRT QA device, the recalculation had a significantly higher sensitivity than any specific device examined. Specificity was lower for the independent recalculation, but overall, incorporating both sensitivity and specificity, ROC analysis found superior performance for the independent recalculation over clinical measurement-based IMRT QA.
Conclusion: A simple independent recalculation using linac class-specific generic beam models had much better sensitivity at detecting unacceptable plans as compared to existing clinical measurement-based IMRT QA methods.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by Public Health Service Grants CA214526 and CA180803 awarded by the National Cancer Institute, United States Department of Health and Human Services