MapCHECK in MapPHAN: QA for Lung SABR

• MapCHECK in MapPHAN is a QA system that integrates a MapCHECK 2 diode array in a solid water phantom for detailed OAR dosimetry in lung SABR.
• It utilizes density overrides, CT characterization, and merged offset measurements to correct angular dependence and CT artifacts for accurate dose comparison.
• ROC analysis with 5%/1 mm and 2%/1 mm criteria demonstrates the method’s sensitivity to clinically significant MLC and isocentre errors in conformal arc treatments.

Searching arXiv for the specified paper to ground the article and confirm citation details. arxiv_search.query({"search_query":"id:(Henry et al., 21 Aug 2025)","max_results":5}) MapCHECK in MapPHAN denotes the use of a MapCHECK 2 diode array housed within the Sun Nuclear MapPHAN solid water case for patient-specific quality assurance of conformal arc lung Stereotactic Ablative Body Radiotherapy, with explicit attention to Organ At Risk dosimetry rather than only target coverage. In the reported implementation, the phantom is modified to accommodate MapCHECK 2, commissioned with density overrides in RayStation to compensate for angular dependence and CT artefacts, and evaluated through gamma analysis and Receiver Operating Characteristic methods against introduced Multi-Leaf Collimator and isocentre errors. The principal conclusion is that, for this QA configuration, a combination of 5% / 1 mm with 95% gamma tolerance and 2% / 1 mm with 90% tolerance provides good sensitivity and specificity for conformal arc SABR QA (Henry et al., 21 Aug 2025).

1. System definition and phantom configuration

The system is built around a MapCHECK 2 (model 1177) diode array comprising 1527 N-type SunPoint diodes with 1 cm detector spacing in a 32 × 26 cm² 2D grid, housed in an acrylic case (Henry et al., 21 Aug 2025). To enable arc QA in a quasi-solid-water geometry, the array is mounted inside the Sun Nuclear MapPHAN solid water case. Because MapPHAN was originally designed for MapCHECK 1 (model 1175), two solid-water inserts measuring 11 × 29 × 300 mm are fabricated to adapt the MapPHAN-MC1 accessory to MapCHECK 2; these inserts are placed on the left and right sides to ensure a tight lateral fit and a contiguous solid-water volume around the detector plane (Henry et al., 21 Aug 2025).

The phantom assembly is CT-scanned in both coronal and sagittal orientations. Substantial streak artefacts are present in the diode plane because of photon starvation, and density overrides are therefore applied in RayStation to every component of the phantom, including MapPHAN slabs, MapCHECK casing halves, the detector region, the couch, and air holes. These overrides are tuned so that the calculated angular dependence of dose matches the measured angular dependence within ±2% for the central diodes, even for lateral beams (Henry et al., 21 Aug 2025). Within the study, this commissioning step is not ancillary; it is described as essential for accurate arc QA because it compensates simultaneously for known angular dependence and CT artefacts.

A common misconception is that insertion of a 2D diode array into a solid-water housing is sufficient for arc-based SABR QA. The reported methodology indicates otherwise: the combination of phantom adaptation, CT characterization in multiple orientations, and explicit density override tuning is integral to the dosimetric validity of the setup (Henry et al., 21 Aug 2025).

2. Conformal arc lung SABR context

The evaluated plans are derived from twenty-three clinical lung SABR cases from Auckland Hospital, all using 6 MV flattened X-ray beams on a Varian Clinac iX with Millennium 120 MLC (Henry et al., 21 Aug 2025). Fourteen patients receive at least 48 Gy to the PTV in 4 fractions, and nine receive at least 60 Gy to the PTV in 8 fractions. Clinical optimisation is performed in Pinnacle v9.8/9.10, after which the plans are anonymised and imported into RayStation v5.0 for error modelling and phantom dose calculation (Henry et al., 21 Aug 2025).

The geometric and planning framework is specified in detail. The ITV is generated from 8-phase respiratory-gated CT acquired with the Varian RPM system, and a 5 mm isotropic margin is added to the ITV to form the PTV. Arc lengths span 180–220°, with 2 or 3 coplanar arcs per plan. Control point spacing is 5° in Pinnacle and is manually edited to 4° for RayStation compatibility (Henry et al., 21 Aug 2025).

The OAR set comprises spinal canal, oesophagus, trachea, proximal bronchial tree, heart, great vessels, ribs/chest wall, brachial plexus, skin, and external normal tissue. Maximum point dose, $D_{\max}$, is the primary OAR metric. For each plan and each OAR, RayStation computes the maximum point dose and then recalculates it after introduction of errors to evaluate dose sensitivity (Henry et al., 21 Aug 2025). This OAR-centric framing is a defining feature of the study. The reported analysis does not treat gamma pass rate as an abstract delivery score; instead, it tests whether gamma behavior tracks clinically relevant OAR dose excursions.

Nine of the twenty-three patients are selected for physical measurements. Their 6 MV conformal arcs are delivered to the MapCHECK-in-MapPHAN assembly on the Varian Clinac iX, with the phantom positioned centrally on the couch and the detector plane aligned to treatment isocentre as in a standard patient setup. The delivered set includes original plans, plans with systematic MLC class-open errors up to 1 mm in this measurement subset, and plans with isocentre shifts that are analyzed primarily in silico while using analysis-software shifts to examine gamma behavior with and without registration correction (Henry et al., 21 Aug 2025).

3. Error modelling and OAR dose sensitivity

Two error classes are investigated: systematic class-open MLC errors and isocentre shift errors. The MLC perturbation is global rather than stochastic: all leaves on both banks are shifted outward, widening the aperture by incremental steps of 0.2 mm. The detailed in-silico OAR analyses extend to 1.0 mm, while the abstract notes that errors up to 2 mm are modelled across the full set of 23 plans (Henry et al., 21 Aug 2025). Python scripts modify Pinnacle plan files before import to RayStation, adjusting MLC positions at all control points, and RayStation recalculates dose using a 0.2 × 0.2 × 0.2 cm³ dose grid for both original and modified plans (Henry et al., 21 Aug 2025).

For each plan and OAR, the dose response to MLC error is quantified through a linear regression slope expressed as

$$\frac{\Delta D_{\text{OAR,max}}}{\Delta e_{\text{MLC}}} \quad \text{(Gy/mm)}.$$

The study reports that OARs close to the PTV, including proximal bronchus, spinal canal, chest wall/ribs, trachea, and brachial plexus, exhibit the largest gradients; their $D_{\max}$ increases steeply even for sub-millimetre MLC widening (Henry et al., 21 Aug 2025). The same logic is used for isocentre shifts, which are modelled as ±1 mm displacements independently along the superior–inferior, right–left, and anterior–posterior axes, again with dose recalculated on the same 0.2 cm grid and OAR-specific Gy/mm gradients extracted from plotted responses (Henry et al., 21 Aug 2025).

Clinically, the study identifies distinct implications for the two error types. A 2 mm systematic MLC class-open error is described as large in SABR because it strongly increases dose spill beyond the target, and the regression lines show that OAR $D_{\max}$ can increase by several Gy per mm of MLC opening depending on OAR proximity (Henry et al., 21 Aug 2025). By contrast, a 1 mm isocentre error shifts the high-gradient distribution relative to adjacent OARs, producing smaller but directionally adverse dose changes, particularly for structures immediately adjacent to the PTV. The reported ROC analysis indicates that 1 mm is near the detection limit for the QA setup from a gamma-metric standpoint (Henry et al., 21 Aug 2025).

This suggests a useful distinction between dosimetric consequence and detectability. The MLC perturbation produces a systematic aperture expansion that is both clinically consequential and gamma-detectable in this framework, whereas the 1 mm isocentre perturbation is clinically relevant but only weakly separable by the measured gamma statistics.

4. Gamma analysis methodology

The comparison between measured and calculated dose uses standard 2D gamma analysis. The gamma index is defined as

$$\gamma(x) = \min_{x'} \sqrt{ \left( \frac{D(x) - D'(x')}{\Delta D} \right)^2 + \left( \frac{\|x - x'\|}{\Delta r} \right)^2 } ,$$

where $D(x)$ is the measured dose at detector position $x$, $D'(x')$ is the planned dose at comparison point $x'$, $\Delta D$ is the dose difference criterion, and $$\frac{\Delta D_{\text{OAR,max}}}{\Delta e_{\text{MLC}}} \quad \text{(Gy/mm)}.$$0 is the distance-to-agreement criterion (Henry et al., 21 Aug 2025).

Six dose-distance criteria are evaluated: 1% / 1 mm, 2% / 1 mm, 2% / 2 mm, 3% / 2 mm, 3% / 3 mm, and 5% / 1 mm. A 10% dose threshold is applied, excluding points below 10% of the maximum dose from gamma analysis (Henry et al., 21 Aug 2025). The study does not fix a clinically acceptable pass-rate threshold a priori; instead, pass-rate thresholds are examined through ROC analysis.

A notable feature of the measurement strategy is the deliberate enhancement of effective sampling resolution. For each plan, two deliveries are acquired: one with the phantom centred at isocentre and a second with the phantom shifted 5 mm superior, corresponding to half the 10 mm diode spacing. The two measurements are then merged in SNC Patient, effectively producing an interpolated detector spacing of about 5 mm and improving sampling in high-gradient regions (Henry et al., 21 Aug 2025). The study describes this as crucial for SABR QA and for sensitivity to small geometric errors.

During analysis, small global shifts are applied to align measured and calculated distributions. For isocentre-error experiments, gamma analysis is performed both with and without these shift corrections to examine the impact of registration (Henry et al., 21 Aug 2025). Reported mean pass rates are lower in coronal orientation than in sagittal orientation, reflecting residual angular dependence and CT artefact effects, although later ROC analysis indicates that the orientation difference does not materially alter discriminative performance (Henry et al., 21 Aug 2025).

A recurrent misunderstanding in SABR QA is that a numerically lenient dose criterion necessarily implies reduced rigor. In this dataset, the proposed 5% / 1 mm criterion pairs a relatively permissive dose-difference allowance with a tight 1 mm DTA and a high gamma pass-rate threshold, so its role is not simply to make plans pass more easily; it is selected because it yields the strongest discrimination between OAR-acceptable and OAR-unacceptable plans in ROC space (Henry et al., 21 Aug 2025).

5. ROC analysis and recommended decision thresholds

The study constructs Receiver Operating Characteristic curves by treating gamma pass rate as a continuous classifier of plan acceptability. Plans are labelled acceptable when OAR doses remain within tolerance based on RayStation $$\frac{\Delta D_{\text{OAR,max}}}{\Delta e_{\text{MLC}}} \quad \text{(Gy/mm)}.$$1 and OAR constraints, and unacceptable when introduced MLC or isocentre errors produce OAR dose exceedance. Sensitivity is defined as the fraction of truly unacceptable plans correctly flagged by gamma failure; specificity is the fraction of truly acceptable plans correctly passing gamma. The Area Under the ROC Curve quantifies discriminative power, and the Youden index identifies the balance point maximizing sensitivity plus specificity minus 1 (Henry et al., 21 Aug 2025).

For MLC class-open errors, the reported AUC values are as follows:

Orientation	Criterion	AUC
Coronal	1% / 1 mm	0.64
Coronal	2% / 1 mm	0.62
Coronal	2% / 2 mm	0.50
Coronal	3% / 2 mm	0.46
Coronal	3% / 3 mm	0.45
Coronal	5% / 1 mm	0.73
Sagittal	1% / 1 mm	0.58
Sagittal	2% / 1 mm	0.62
Sagittal	2% / 2 mm	0.61
Sagittal	3% / 2 mm	0.70
Sagittal	3% / 3 mm	0.62
Sagittal	5% / 1 mm	0.78

These results place the MLC-error discrimination of MapCHECK in MapPHAN in the range of approximately 0.6–0.8 AUC, with 5% / 1 mm yielding the highest AUC in both orientations (Henry et al., 21 Aug 2025). The study therefore identifies 5% / 1 mm as the best-performing single gamma criterion for detection of clinically relevant OAR dose violations caused by MLC class-open errors. It also notes that the orientation dependence is modest and likely statistical rather than systematic (Henry et al., 21 Aug 2025).

For isocentre shifts, the reported AUCs are lower: 0.15–0.42 in coronal orientation and 0.32–0.67 in sagittal orientation, with the best sagittal values at 2% / 1 mm and 5% / 1 mm (Henry et al., 21 Aug 2025). The interpretation given is twofold. First, a 1 mm isocentre shift is small relative to measurement resolution and phantom uncertainties. Second, current clinical 1 mm isocentre tolerances appear conservative, since many 1 mm shift plans do not violate OAR thresholds and therefore are not consistently expected to fail gamma (Henry et al., 21 Aug 2025).

From ROC curves and visual assessment of the Youden index, the study recommends a dual-criterion strategy for conformal arc SABR QA: a primary criterion of 5% / 1 mm with a 95% gamma pass-rate tolerance, and a secondary stricter criterion of 2% / 1 mm with a 90% gamma pass-rate tolerance, with 80–90% noted as a possible departmental range for the latter (Henry et al., 21 Aug 2025). The rationale is explicit: the first criterion provides a strong balance of sensitivity and specificity in steep-gradient SABR geometry, while the second increases sensitivity to subtler OAR overdose risks.

6. OAR-centric interpretation, workflow implications, and limitations

The OAR dosimetry results show that proximity to the PTV dominates error sensitivity. OARs lying near the PTV edge, such as proximal bronchus, spinal canal, chest wall/ribs, trachea, and brachial plexus, have the largest Gy/mm dose gradients, and even 0.5–1.0 mm MLC aperture expansion measurably increases their $$\frac{\Delta D_{\text{OAR,max}}}{\Delta e_{\text{MLC}}} \quad \text{(Gy/mm)}.$$2 and can move them across tolerance thresholds (Henry et al., 21 Aug 2025). More peripheral OARs and external tissue have lower gradients, although they still show dose increases consistent with increased intermediate-dose spill. For isocentre shifts, the dose changes at 1 mm are smaller but remain relevant for near-PTV OARs, reinforcing the importance of precise image guidance and couch setup in clinical SABR (Henry et al., 21 Aug 2025).

The paper further reports that plans whose OAR $$\frac{\Delta D_{\text{OAR,max}}}{\Delta e_{\text{MLC}}} \quad \text{(Gy/mm)}.$$3 exceeds tolerance tend to have lower gamma pass rates, whereas plans remaining within tolerance show high pass rates. In this sense, gamma is not treated merely as a geometric congruence metric but as a surrogate classifier for OAR protection performance (Henry et al., 21 Aug 2025). This is a significant conceptual shift from prior QA paradigms that emphasize PTV coverage. Here, the decisive clinical question is whether the QA metric predicts violation of OAR constraints in circumstances where PTV dose may already be intentionally compromised to protect nearby normal structures.

For implementation in a medical physics department, the reported workflow comprises CT imaging of the MapCHECK-in-MapPHAN assembly in both coronal and sagittal orientations; application of density overrides to the phantom, detector region, air cavities, and couch; optimization of these overrides until measured and planned angular dose agree within ±2% for central diodes; two measurements per plan, one at isocentre and one with a 5 mm superior shift; merging of the measurements in SNC Patient; use of a 10% dose threshold; and small global alignment shifts during analysis, with optional omission of registration correction when specifically probing isocentre sensitivity (Henry et al., 21 Aug 2025). The recommended routine QA criteria are 5% / 1 mm at 95% pass rate as the primary standard and 2% / 1 mm at 90% pass rate as a stricter companion standard (Henry et al., 21 Aug 2025).

The limitations stated in the study are principally spatial and angular. Even after merging two offset measurements, the effective sampling is 5 mm, which can limit sensitivity to very fine spatial errors and extremely sharp gradients. MapCHECK’s native 1 cm pitch means that the highest gradient regions are under-sampled relative to film or high-resolution 3D dosimetry, although the study concludes that SABR arc plans in this dataset were still adequately assessed. Angular dependence requires careful commissioning and density overrides, and residual differences between coronal and sagittal measurements persist, even though they do not significantly alter ROC performance (Henry et al., 21 Aug 2025). The paper notes that film remains the highest-resolution option but is time-consuming, whereas MapCHECK/MapPHAN provides a practical, time-efficient alternative with instantaneous readout (Henry et al., 21 Aug 2025).

Taken together, the reported evidence characterizes MapCHECK in MapPHAN as an OAR-relevant QA platform for conformal arc lung SABR when it is used with deliberate commissioning, merged offset measurements, and gamma criteria tailored to steep-dose-gradient geometry. Its strongest demonstrated capability is the detection of clinically meaningful systematic MLC class-open errors through gamma pass rates that correlate with OAR $$\frac{\Delta D_{\text{OAR,max}}}{\Delta e_{\text{MLC}}} \quad \text{(Gy/mm)}.$$4 threshold breaches, while 1 mm isocentre errors appear to lie close to the dosimetric detection limit of the configuration (Henry et al., 21 Aug 2025).