Volume threshold of brain metastases for Linac-based SRS
For2-3 brain metastases, SRS candidates, GK vs linac VMAT planning
TL;DRIn a GK-vs-VMAT replanning study, GK beat VMAT on plan quality for targets ≤0.4 cc, while VMAT won for targets >0.4 cc.
HyperArc shrinks the GK advantage on small targets: for lesions ≤0.4 cc, NT V12Gy was 0.79 vs GK 0.70 (p=0.1923), while SmartArc was clearly worse at 1.25 (p=0.0019). The cutoff depends on the platform, and VMAT beam-on time was 8.33 to 13.58 min vs GK's 56.25.
For a pt with 2-3 brain mets where every target is >0.4 cc, this supports VMAT-based linac SRS as dosimetrically equal or better than GK with shorter beam-on time. It does not settle platform choice for sub-0.4 cc lesions or for >3 targets.
HyperArc shrinks the GK advantage on small targets: for lesions ≤0.4 cc, NT V12Gy was 0.79 vs GK 0.70 (p=0.1923), while SmartArc was clearly worse at 1.25 (p=0.0019). The cutoff depends on the platform, and VMAT beam-on time was 8.33 to 13.58 min vs GK's 56.25.
12 details
Retrospective treatment-planning comparison at one institution. 26 pts, 76 lesions, each GK plan replanned with SmartArc and HyperArc. ROC analysis on RTOG CI and NT V12Gy set the threshold volumes.
Pts treated with GK SRS for brain metastases in 2018-2019 who had two or three targets (26 selected from 121). Targets with a prescription below 12 Gy were excluded from the V12Gy analysis.
Linac plans used HD-120 MLC with FFF beams, with one or two isocenters per pt. Example prescriptions ranged from 13 Gy to 24 Gy single fraction.
The ROC thresholds were 0.4 cc for SmartArc (CI and V12Gy), 0.2 cc for HyperArc CI and 0.5 cc for HyperArc V12Gy. Averaged, they give the proposed 0.4 cc cutoff: GK is favored at or below it and VMAT above it. Median beam-on time was 56.25 min for GK vs 8.33 and 13.58 min for VMAT.
| Plan · metric | Threshold | Sensitivity | Specificity | Accuracy |
|---|---|---|---|---|
| SmartArc · RTOG CI | 0.4 cc | 81.8% | 87.8% | 85.9% |
| HyperArc · RTOG CI | 0.2 cc | 100.0% | 79.7% | 81.7% |
| SmartArc · NT V12Gy | 0.4 cc | 79.2% | 94.4% | 88.3% |
| HyperArc · NT V12Gy | 0.5 cc | 89.5% | 82.9% | 85.0% |
| Metric | GK | SmartArc (p) | HyperArc (p) |
|---|---|---|---|
| ≤0.4 cc · RTOG CI | 1.68 | 1.85 (0.0013) | 1.46 (0.032) |
| ≤0.4 cc · NT V12Gy (cc) | 0.70 | 1.25 (0.0019) | 0.79 (0.1923) |
| ≤0.4 cc · GI | 3.75 | 6.62 (0.0041) | 4.96 (0.0284) |
| ≤0.4 cc · V4.5Gy (cc) | 2.58 | 6.82 (0.0050) | 6.26 (0.0065) |
| >0.4 cc · RTOG CI | 1.59 | 1.36 (<0.0001) | 1.23 (<0.0001) |
| >0.4 cc · GI | 3.19 | 3.30 (0.5297) | 3.30 (0.2969) |
| >0.4 cc · V4.5Gy (cc) | 17.55 | 16.04 (0.0152) | 13.52 (<0.0001) |
This is consistent with Liu et al, where VMAT matched or beat GK at a median volume of 0.5 cc. Potrebko et al and Vergalasova et al split targets by mean volume or diameter bins and planned more targets per pt (median eight and seven), so their cutoffs are not directly comparable.
The cutoff is an average of four ROC thresholds, not tested in a validation set. For ≤0.4 cc, Table 4 shows HyperArc CI 1.46 vs GK 1.68 (p=0.032), which conflicts with the text's 'comparable conformity'. Closely adjacent targets were dropped, and these are exactly the cases where MLC dose bridging matters most.
The useful signal is that the threshold depends on the platform: HyperArc stays competitive with GK on small-target V12Gy where SmartArc does not. No clinical endpoint links these dosimetric differences to radionecrosis or local control.
Single-institution retrospective replanning with dosimetric surrogates only. The threshold is an unvalidated average of ROC cutoffs, with no clinical toxicity or control endpoint.
- Do sub-0.4 cc dosimetric differences translate into radionecrosis or local control differences?
- Does the threshold hold with more than three targets per patient?
- Is the averaged 0.4 cc cutoff reproducible in an independent validation cohort?
📚 Sources · 📄 1 paper
The longer read
The question this paper asks is practical: as linac VMAT radiosurgery displaces dedicated Gamma Knife units in many departments, which lesions still justify the GK workflow? The answer offered, a 0.4 cc cutoff, is clean and easy to remember. It deserves less confidence than its precision suggests. The number is not a single ROC-optimal point. It is the average of four thresholds (0.4, 0.2, 0.4 and 0.5 cc) from two planning systems and two metrics, and it was never tested on a held-out set. The individual thresholds are the more honest result, and they say the crossover depends on the platform. With HyperArc, conformity crossed over at 0.2 cc, and small-target NT V12Gy was not significantly different from GK (0.79 vs 0.70, p=0.1923). With SmartArc, GK was clearly better on every metric below the cutoff.
That distinction matters more than the headline. A department running a modern single-isocenter multi-target solution should read this as evidence that the small-lesion gap is narrow and mostly confined to low-dose spread (V4.5Gy and GI). A department running a generic VMAT optimizer should read it as a reason to keep sub-0.4 cc lesions on a sharper-penumbra system when one is available. There is also a reporting inconsistency to note. Table 4 shows HyperArc RTOG CI 1.46 vs GK 1.68 (p=0.032) for small targets, which reads as HyperArc being more conformal, yet the text calls conformity comparable. The table may define CI differently or the columns may be transposed, but a reader cannot resolve it from the source.
Against the prior literature the direction is consistent. Liu et al reported VMAT matching or beating GK at a median volume of 0.5 cc. Vergalasova et al found VMAT conformity better than GK above roughly 1 cm diameter. Potrebko et al found a V12Gy advantage for GK in their small-target group. The present cohort had fewer targets per pt (median three, vs eight and seven in those series), and the authors acknowledge that MLC leakage between targets worsens with more lesions. The 0.4 cc cutoff should therefore be read as a best case for VMAT, not something that carries over to 5-10 lesion cases.
The larger gap is clinical. Every endpoint here is dosimetric. NT V12Gy is a reasonable radionecrosis surrogate, but the absolute small-target differences (fractions of a cc) are unlikely to separate pts on toxicity, and none of these plans was delivered. Excluding closely adjacent targets removes exactly the geometry where dose bridging would penalize VMAT most. What would make the conclusion wrong is a validation set in which the crossover shifts substantially, or clinical data showing that sub-cc V12Gy differences of this size do not change necrosis rates. The 56 vs 8 to 14 min beam-on difference is the most robust finding in the paper, and for many pts it will weigh more than a small conformity edge on a tiny lesion.