Premium
SU‐G‐BRB‐16: Vulnerabilities in the Gamma Metric
Author(s) -
Neal B,
Siebers J
Publication year - 2016
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4956923
Subject(s) - computer science , metric (unit) , imaging phantom , image quality , quality assurance , field (mathematics) , set (abstract data type) , algorithm , artificial intelligence , mathematics , image (mathematics) , optics , physics , operations management , economy , pure mathematics , economics , programming language , service (business)
Purpose: To explore vulnerabilities in the gamma index metric that undermine its wide use as a radiation therapy quality assurance tool. Methods: 2D test field pairs (images) are created specifically to achieve high gamma passing rates, but to also include gross errors by exploiting the distance‐to‐agreement and percent‐passing components of the metric. The first set has no requirement of clinical practicality, but is intended to expose vulnerabilities. The second set exposes clinically realistic vulnerabilities. To circumvent limitations inherent to user‐specific tuning of prediction algorithms to match measurements, digital test cases are manually constructed, thereby mimicking high‐quality image prediction. Results: With a 3 mm distance‐to‐agreement metric, changing field size by ±6 mm results in a gamma passing rate over 99%. For a uniform field, a lattice of passing points spaced 5 mm apart results in a passing rate of 100%. Exploiting the percent‐passing component, a 10×10 cm 2 field can have a 95% passing rate when an 8 cm 2 =2.8×2.8 cm 2 highly out‐of‐tolerance (e.g. zero dose) square is missing from the comparison image. For clinically realistic vulnerabilities, an arc plan for which a 2D image is created can have a >95% passing rate solely due to agreement in the lateral spillage, with the failing 5% in the critical target region. A field with an integrated boost (e.g whole brain plus small metastases) could neglect the metastases entirely, yet still pass with a 95% threshold. All the failure modes described would be visually apparent on a gamma‐map image. Conclusion: The %gamma<1 metric has significant vulnerabilities. High passing rates can obscure critical faults in hypothetical and delivered radiation doses. Great caution should be used with gamma as a QA metric; users should inspect the gamma‐map. Visual analysis of gamma‐maps may be impractical for cine acquisition.