Optimization of Gamma Parameters for EPID-Based Pretreatment Detection of Delivery Errors in IMRT Fields
J Gordon1*, J Siebers2, (1) Henry Ford Health System, Detroit, MI, (2) Virginia Commonwealth University, Richmond, VATH-A-BRA-7 Thursday 8:00:00 AM - 9:55:00 AM Room: Ballroom A
Purpose: Gamma analysis is widely used for intensity modulated radiation therapy (IMRT) quality assurance. However, when calculated on a discrete grid with conventional dose/distance criteria of f_dose = 3%/f_dist = 3mm, gamma exhibits poor correlation with dose errors and poor specificity/sensitivity for detecting fluence deviations. This work shows how gamma criteria may be optimized for error detection, and uses repeat images of 11 clinical IMRT fields to quantify delivery errors that can be reliably detected via optimal analysis of electronic portal (EPID) images.
Methods: Multiple EPID images were acquired of 11 clinical IMRT fields. For each field, images were aligned and averaged, generating a reference image. Gamma was computed for individual images relative to reference using two Methods: discrete grid (0.37mm^2 pixel size) and continuous (interpolation-free). Rectangular errors were inserted into measured images, and receiver operating characteristic (ROC) studies determined parameters that yield maximum image classification accuracy. Classification was deemed reliable if =95% of images were correctly classified.
Results: Key parameters are f_grad=0.37xf_dose/f_dist and the gamma distribution tail cutoffs used to classify images as errored or error-free. f_grad controls the 'looseness' of the distance-to-agreement (DTA) search. Classification is optimal when f_grad is loose enough to ignore normal image-to-image variations, but tight enough to detect delivery errors. Conventional 3%/3mm criteria give sub-optimal error detection. Even with optimal parameters, calculation on a discrete grid introduces sufficient sampling errors as to seriously degrade error detection. If continuous gamma calculation is employed with optimal parameters, one can reliably detect small errors, e.g., dose deviations of 5-10% over areas of 3x3 pixels (~1mm^2) at isocenter.
Conclusions: Optimal detection of IMRT delivery errors using EPID images is achieved with continuous gamma, and optimal criteria (f_dose, f_dist and classification cutoffs). Caution must be exercised when gamma is calculated on a discrete grid using conventional 3%/3mm criteria.