IMRT patient-specific quality assurance — the process of verifying that a treatment plan delivers dose to a phantom as predicted before delivering it to the patient — is among the most resource-intensive QA activities in a radiation oncology physics department. For a department treating fifteen to twenty IMRT patients per day, patient-specific QA consumes a substantial fraction of physicist and linac time. The metric at the center of every IMRT QA session is the gamma passing rate: a composite measure of dose magnitude agreement and spatial accuracy between measured and calculated dose distributions. Understanding what gamma passing rates actually tell you, and where measurement uncertainty limits their interpretive value, is essential for establishing clinically meaningful action thresholds.
The Gamma Analysis Framework
Gamma analysis, introduced by Low et al. in the late 1990s, evaluates each measurement point against its corresponding calculated dose point using a combined criterion: dose difference tolerance (DTA in percent) and distance-to-agreement tolerance (DTA in mm). A measurement point passes if there exists a calculated point within the DTA spatial radius where the dose agrees within the DD percentage. The composite gamma value at each point is the minimum normalized distance to a passing calculated point; gamma <1 indicates a passing point.
Clinical practice has converged on 3%/3mm as a common gamma criterion for IMRT QA, though 2%/2mm and 3%/2mm criteria are increasingly used as detector resolution has improved. AAPM Task Group 218 (published 2018) established action thresholds and tolerance limits for IMRT QA gamma analysis: a 3%/3mm global gamma passing rate of ≥95% as a tolerance limit, with 90% as an action threshold below which investigation is required. These guidelines have been adopted as reference standards by Japanese clinical physics practice through JASTRO and the Japan Society of Medical Physics (JSMP) recommendations.
Sources of Measurement Uncertainty in IMRT QA
The clinical interpretation of gamma passing rates is complicated by multiple sources of measurement uncertainty that are often underestimated in routine QA practice:
Detector resolution and volume averaging: 2D detector arrays used for IMRT QA have spatial resolution limited by detector pitch — typically 5–10 mm for ion chamber arrays, 2.5–5 mm for diode arrays. High-dose-gradient regions in IMRT plans, particularly at field edges and around OAR dose falloff areas, are susceptible to volume averaging artifacts where the detector's finite size averages the dose across a gradient, systematically misrepresenting the measured distribution. Using a 3mm DTA criterion with a 7.5mm-pitch detector is not a conservative combination; the DTA criterion can mask spatial inaccuracies that the detector physically cannot resolve.
Phantom geometry effects: IMRT QA performed in a flat slab or cylindrical phantom delivers dose in a geometry that differs from the patient geometry. The MLC leaf positions, gantry angles, and dose rates are the same, but the tissue heterogeneities, the scatter geometry, and the absolute dose values at the measurement plane differ from patient conditions. A plan that passes gamma QA in a flat phantom can exhibit dosimetric deviations in patient anatomy that the phantom measurement does not capture — particularly for highly modulated plans with significant lateral electron transport effects.
Linac output variability: Day-to-day linac output variation — typically within 1% for well-maintained machines — contributes to gamma passing rate variability independently of plan quality. A QA measurement performed on a day when the linac output is at the low end of its tolerance band will yield systematically lower dose readings, potentially failing gamma criteria on a plan that would pass on average-output days. Departments that do not normalize for daily output in their QA analysis workflow introduce output variability as a confound in their gamma passing rate trends.
Japan Clinical Physics Standards and Action Thresholds
JSMP and JASTRO recommendations on IMRT QA are broadly aligned with AAPM TG-218 but include Japan-specific guidance on detector calibration traceability requirements (calibration against National Metrology Institute of Japan standards via secondary standard dosimetry laboratories) and minimum measurement frequency requirements for different plan modulation levels.
For highly modulated plans — those with high modulation factors, often encountered in head-and-neck and gynecological IMRT — JASTRO recommends supplementing 2D detector array measurements with point dose measurements in high-dose regions to cross-check the array result. This two-measurement approach provides a partially independent check that catches systematic detector errors (e.g., a chamber array where a specific row of detectors has calibration drift) that would not be apparent from the passing rate alone.
At an academic cancer center running approximately thirty IMRT QA measurements per week, implementing a systematic point-dose cross-check for plans with modulation factors above 3.0 identified two cases over a six-month period where the 2D array reported ≥95% passing at 3%/3mm but the point dose measurement showed a 4.5% deviation in the high-dose region. Both cases were re-planned with reduced modulation before treatment. The 2D array measurement alone would not have triggered investigation; the combined protocol caught deviations that the single-metric threshold missed.
When Gamma Passing Rate Is and Is Not Informative
The fundamental limitation of gamma passing rate as a clinical action criterion is that it is sensitive to the wrong things and insensitive to others. A high gamma passing rate in the low-dose penumbra regions of a plan — where absolute dose is low — contributes many passing points to the rate without being clinically relevant. Meanwhile, a small systematic error in the high-dose region near the tumor can depress the passing rate by only a few percentage points while representing a clinically significant dose inaccuracy.
We are not arguing that gamma analysis should be abandoned — it remains the practical standard for IMRT QA and has real clinical utility when interpreted correctly. The argument is that gamma passing rate is a necessary but not sufficient indicator of plan delivery accuracy. The AAPM TG-218 recommendation to investigate clinical significance for failed QA cases explicitly acknowledges this: a plan that fails gamma criteria in a low-dose peripheral region has different clinical implications than one that fails in the center of the GTV.
Normalization choice also affects interpretability significantly. Global normalization (dividing all dose differences by the maximum dose in the field) penalizes low-dose region deviations less severely than local normalization (dividing by the local dose at each point). The 3%/3mm global criterion is more lenient in low-dose regions than 3%/3mm local; plans that appear to comfortably pass on global criteria may fail local criteria in clinically relevant regions. Departments should document which normalization scheme they use and ensure that action thresholds are defined consistently with that scheme.
How AI-Assisted Planning Affects QA Outcomes
An underexplored relationship in the AI planning literature is the effect of AI-generated plan modulation on IMRT QA passing rates. AI-assisted planners, when trained on historical plans from experienced physicists, tend to produce plans with modulation levels consistent with that training distribution. If the training plans were developed to meet both clinical DVH criteria and practical deliverability criteria — a common implicit constraint in experienced physicist planning — the AI-generated plans may have lower average modulation factors than plans generated by less experienced planners applying tight constraints without deliverability awareness.
Lower plan modulation generally correlates with higher IMRT QA passing rates, because highly modulated fluence maps are the primary source of dose-gradient regions that challenge gamma analysis. This is not a reason to plan with artificially low modulation — clinical DVH objectives should drive modulation level, not QA passability — but it is a mechanism by which AI planning can indirectly improve QA outcomes by producing plans that achieve clinical objectives with appropriate rather than excessive modulation.
The clinical workflow implication is that departments using AI-assisted planning should track IMRT QA passing rate distributions before and after AI implementation as part of their clinical performance monitoring. A systematic shift toward higher passing rates, correlated with reduced plan modulation on comparable case types, is evidence that the AI planning workflow is producing more deliverable plans — a clinically meaningful outcome beyond DVH metrics alone.