SU‐GG‐T‐195: On the Correlation of Conventional IMRT QA Metrics to Clinical Impact in Per‐Patient Dose QA

Abstract

Purpose: Per‐patient IMRT QA is often performed by comparing measured planar phantom dose to calculated planar phantom dose, per beam. The most common analysis metric is percent passing 3%/3mm (DTA). Similarly, TG‐119 reports utilizes a metric of 3%/3mm (Gamma) when discussing IMRT commissioning. In this work, we analyze the correlation of conventional per‐beam IMRT QA metrics to: 1) patient dose comparisons (3D %diff/DTA), and 2) important clinical dose endpoints in volumetric regions‐of‐interest. Method and Materials: For 24 Head/Neck IMRT plans, an accurate TPS beam model was error‐induced (four unique ways) to simulate IMRT calculation and/or delivery errors. The error‐free beam models were used to generate “virtual planar QA measurements” and analyzed in conventional ways against the error‐induced beams. The error‐free 3D patient doses were compared against the error‐induced 3D patient doses. This method produced a robust simulation (high precision, high data density of 3D patient dose) which is necessary given the practical inability to measure dose inside complicated patient anatomy. The 96 error‐induced plans were compared to the error‐free plans to quantify the correlation of conventional Planar QA metrics with relevant clinical endpoints such as: max cord dose, mean parotid dose, and D95 of the principal CTV. Results: No correlation was found between 3%/3mm conventional IMRT QA criteria and all clinical endpoints studied (R‐squared values of: 0.0314, 0.0272, 0.0899, and 0.0248 for predicting errors in the cord max, left parotid mean, right parotid mean, and CTV D95). There was also no correlation when considering 2%/2mm and 1%/1mm metrics. In addition, there were many examples of both false positives and false negatives using conventional IMRT QA methods/metrics to predict accurate patient dose. Conclusion: The location and magnitude of the errors was far more important than the quantity of the errors, and moving to patient ROI‐based error simulations is strongly recommended.