Interinstitutional plan quality: Elements MME vs RapidArc single-isocenter SRS
For4-10 brain metastases, single-isocenter linac SRS, single fraction
TL;DRRapidArc lowered median V12Gy 19.2 vs 23.7 cc (P<.0001) and improved conformity; MME lowered V5Gy and mean brain dose (2.57 vs 2.76 Gy).
The platforms trade dose bands rather than one beating the other. RA lowers the V12Gy/V8Gy band tied to radionecrosis, while MME lowers dose below roughly 6 Gy. The V12 advantage holds only above 2.1 cc total PTV (ratio mean 1.24); below that, MME matched or beat RA on V12 in 4 of 6 cases.
For a pt with 4-10 brain mets and total PTV above 2.1 cc, this favors single-isocenter VMAT for lower V12Gy. With very small total PTV, MME matched or beat RA on V12 in 4 of 6 cases. Low-dose spill differences have no outcome correlation yet.
The platforms trade dose bands rather than one beating the other. RA lowers the V12Gy/V8Gy band tied to radionecrosis, while MME lowers dose below roughly 6 Gy. The V12 advantage holds only above 2.1 cc total PTV (ratio mean 1.24); below that, MME matched or beat RA on V12 in 4 of 6 cases.
16 details 3 trials watching
Paired dosimetric replanning study at 2 institutions (TJU, UAB). 15 previously treated pts from each site were replanned by the other site with its own technique. Comparison by paired 2-tailed Wilcoxon signed-rank, prespecified α .05.
30 pts with 4 to 10 brain metastases, 217 targets, median 7.5 per case. Individual target volume 0.014 to 17.73 cm3; total PTV 0.49 to 27.32 cm3, median 7.05 cm3.
14 to 24 Gy single fraction, prescriptions unchanged between plans and largely RTOG 9508-based. RA used 2 to 4 noncoplanar VMAT arcs at 10 MV FFF with 3 ring tuning structures. MME used automated noncoplanar DCAs at 6 MV FFF with up to 1 mm MLC margin and island-blocking avoidance.
Per target: RTOG CI, Paddick CI, coverage. Per plan: V12Gy, V8Gy, V5Gy, mean brain dose, plus arc, MU, and beam-on estimates.
RA favored conformity and the V12Gy/V8Gy band; MME favored V5Gy and Dmean. The brain DVHs cross near 6 Gy, and below 2.1 cc total PTV MME was favorable on most metrics.
| Metric | MME (DCA) | RA (VMAT) | p |
|---|---|---|---|
| RTOG CI (per target) | 1.38 | 1.23 (abstract: 1.21) | <.0001 |
| Paddick CI (per target) | 0.70 | 0.77 | <.0001 |
| V12Gy (cc) | 23.7 | 19.2 | <.0001 |
| V8Gy (cc) | 53.6 | 44.1 | .024 |
| V5Gy (cc) | 141.4 | 142.8 | .009 |
| Mean brain dose (Gy) | 2.57 | 2.76 | <.0001 |
| Mean target coverage | 97.5% | 96.9% | .007 |
Gevaert et al (10 cases, 40 mets) found higher V12Gy and V5Gy for RA. Its Paddick CI of 0.65 ± 0.08 (MME) and 0.67 ± 0.16 (RA) sits below this study's 0.70 and 0.77, which the authors read as plans that were not fully optimized. Narayanasamy et al (8 pts, 40 targets) found the platforms comparable but was likely underpowered.
Different beam energies and dose algorithms (pencil beam vs AAA) per platform, so platform effects cannot be cleanly separated from energy and algorithm. Each platform had one expert institution, so planner skill is part of the RA result. Hotspots and homogeneity were not compared, and text and Table 1 medians disagree for V12.
Both platforms met clinical goals, and the differences are small and dose-band specific. Moderate isodose spill has retrospective links to radionecrosis, while low-dose spill has no toxicity correlation yet. It may matter more as pts survive to repeat SRS or WBRT.
Dosimetric planning study with no clinical endpoints; 6 vs 10 MV energy and two dose algorithms confound the platform comparison. Platform and planner effects not separable.
- Does low-dose brain spill predict cognitive decline after repeat SRS? n=80 · primary completion 2026-06 · prospective neurocognitive testing mapped after SRSn=110 · primary completion 2026-12 · MRI biomarkers of NCF decline after SRS vs WBRTn=90 · primary completion 2028-12 · cognitive-sparing SRS of parcellated critical regions
- Does HyperArc or newer Elements shift the dosimetric tradeoff?
- How does non-expert VMAT planning compare with automated MME?
📚 Sources · 📄 1 paper
The longer read
The useful frame for this comparison is that neither platform dominates; each wins a different part of the dose-volume curve, and the crossover sits near 6 Gy. That matters because the two bands carry very different levels of evidence. The moderate isodose region (V8 to V12) has a retrospective literature linking it to radionecrosis, so RA's lower V12Gy is the more clinically interpretable advantage. MME's edge sits in the low-dose bath, where the authors themselves acknowledge no toxicity correlation exists. The case for caring about low-dose spill rests on an argument about cumulative dose in long survivors receiving serial salvage courses, which is plausible but untested.
Against the two earlier head-to-head comparisons, this one is larger and better powered, and it reverses part of the Gevaert finding that RA also carried more low-dose spill. The authors attribute the earlier results to plans that were not fully optimized. That claim is supported by the Paddick CI gap, but it also reveals the central methodological problem: the result for VMAT depends heavily on who builds the plan. Every RA plan here came from one expert planner using the institution's own published technique. The RA arm is therefore closer to a best-case benchmark than to a typical community VMAT plan, while MME's automated templates are likely more reproducible across centers. A reader at a center without that VMAT expertise should expect RA's conformity and V12 advantage to shrink.
The second issue is confounding by hardware. MME ran at 6 MV FFF with a pencil-beam calculation, and RA ran at 10 MV FFF with AAA. Exporting all plans to MIM removes DVH-calculation differences, but it cannot remove differences already in each system's dose calculation, especially for very small targets near the field edge. The authors suggest penumbra shape may explain part of MME's low-dose advantage in small-volume cases. That means the volume-dependent finding, where MME is favorable when total PTV is under 2.1 cc, rests on 6 cases and is the result most exposed to this confound.
The effect sizes also deserve scrutiny before they drive platform decisions. The V5Gy medians differ by about a cubic centimetre despite P=.009, while the Table 1 means show a wider gap, so the distribution is skewed and a few cases drive the difference. Reporting discrepancies between the abstract, text, and table (RA RTOG CI 1.21 vs 1.23; MME V12 median 23.7 vs 24.0) do not change the direction but do lower confidence in the exact values.
What this does not settle is whether any of these differences changes outcomes. A prospective cohort linking platform-specific V12 and low-dose volumes to radionecrosis and neurocognition would. So would a replication with modern optimizers (HyperArc, newer Elements) using the shared de-identified dataset the authors offer. Until then, the reasonable reading is that both platforms are adequate. Expert VMAT buys a modest reduction in V12, and automated DCA buys consistency and a lower low-dose bath.