Inappropriate SRS fractionation

[RadOncDoc21]

EMT409, you're discussing single isocenter management of multiple lesions? On one plan? Using single fraction? Even for Linac-based SRS that isn't overtly common, right?. Only center I am aware of routinely doing single isocenter and then treating brain mets across the entirety of brain with single fraction are UAB and that's because they have hyperARC (which looks cool FWIW). Even with linac-based, we don't routinely do single iso, multiple lesions (unless in close proximity to one another), single fraction.

Same dose also right?

I’ve seen hyperARC and it does look awesome. Right now if we do it, it’s usually close lesions, similar size and dose.

---

Members don't see this ad.

 

[scarbrtj]

For this to be just semantics, and equivalent to an IDL based prescription, the heterogeneity in each target would have to be identical, which it never is.

it would be fair to say, perhaps, lesion A and B got 18 Gy at the 60% IDL and C,D,E got 18 Gy at the 80% IDL in a multi-target plan (that tells me heterogeneities are different, and that the max point dose inside A/B is 30 Gy, and 22.5 Gy inside C/D/E). Sometimes maybe A is in the pons and it's pretty homogenously dosed. That might be: A got 14 Gy at the 90%, B 18 Gy at the 60%, and C/D/E 18 Gy at the 80%; almost thinking about the single iso plan as three separate plans (when of course they're not separate). This is what I might "write" in the plan summary. If this is frowned on, I'd like to know why. I think another way is to state lesional min/max doses. Although, being a romanticist/nostalgist, I still like the "old lingo" of SRS. Either way, if I see a plan summary, I'd like to know lesional min/maxes, and an idea of the plan heterogeneity(ies).

 

[scarbrtj]

Even for Linac-based SRS that isn't overtly common, right?

It should be common. It's probably not. But anybody who has RapidArc (which to me implies a modern machine) "should" be doing it, when they can. You don't need HyperArc for that.
EDIT: sometimes, static multi-field IMRT is easier to plan for some configurations vs RapidArc. So even RapidArc is not a necessity per se, IMHO.

 

[nkmiami]

EMT409, you're discussing single isocenter management of multiple lesions? On one plan? Using single fraction? Even for Linac-based SRS that isn't overtly common, right?. Only center I am aware of routinely doing single isocenter and then treating brain mets across the entirety of brain with single fraction are UAB and that's because they have hyperARC (which looks cool FWIW). Even with linac-based, we don't routinely do single iso, multiple lesions (unless in close proximity to one another), single fraction.

We are a community hospital who have been using the UAB single isocenter vmat technique for 2+ years, no hyperarc, but I almost aways fractionate to "accomodate" margin. It has been quite streamlined and I rarely give whole brain.
Adding margin, at the end of the day is a philosophy. For the vast majority of patients, systemic therapy confounds marginal misses. On average, these pts live 3 to 4 months on national protocols- because they failed systemic therapy- too little time to assess lesion control- yet this makes up a large amount of the area under the local control curve in many series. When they live 1 year+, it is mostly due to a strong contribution from systemic therapy, which very well may also offset marginal dose inadequacies.
In the absence of a drug, 6 Gy x5 extracranially would not provide 90% local control for almost any tumor type.
edit:
I guess you could say the same about 15 Gy vs 22 Gy, but the 15 Gy was for larger tumors.

 

[emt409]

EMT409, you're discussing single isocenter management of multiple lesions? On one plan? Using single fraction? Even for Linac-based SRS that isn't overtly common, right?. Only center I am aware of routinely doing single isocenter and then treating brain mets across the entirety of brain with single fraction are UAB and that's because they have hyperARC (which looks cool FWIW). Even with linac-based, we don't routinely do single iso, multiple lesions (unless in close proximity to one another), single fraction.

UAB is my center, and actually the HyperArc validation was my post-doc project in concert with the Varian developers. We've been doing single-isocenter, multi-met SRS for about 6 or 7 years now. HyperArc uses a modified version of our recipe to streamline the planning process, and also adds several other features including automated collision detection, collimator angle optimization, and the option of automated MV waypoint imaging, etc.

You can absolutely do high quality multi met SRS without HA, but HA just makes it much easier, and faster to plan. It's dramatically reduced the turnaround time for SRS plans, much more than I thought it would TBH.

Prior to HA, at least 100 centers worldwide were using the single-isocenter recipe, so I think it's more widespread than is realized. If people want, I'll post the link to our recipe here.

I think we probably are willing to treat higher numbers of mets with the technique than most centers, just because of the comfort we've gained through our long experience, but we still WBRT a not-insignificant number of patients, particularly if we're worried about LMD. Even in our own center, we still get groans in peer-review when a single-fraction SRS plan to 20 mets comes on the board. But the plans are good, and so is the control rate.

it would be fair to say, perhaps, lesion A and B got 18 Gy at the 60% IDL and C,D,E got 18 Gy at the 80% IDL in a multi-target plan (that tells me heterogeneities are different, and that the max point dose inside A/B is 30 Gy, and 22.5 Gy inside C/D/E). Sometimes maybe A is in the pons and it's pretty homogenously dosed. That might be: A got 14 Gy at the 90%, B 18 Gy at the 60%, and C/D/E 18 Gy at the 80%; almost thinking about the single iso plan as three separate plans (when of course they're not separate). This is what I might "write" in the plan summary. If this is frowned on, I'd like to know why. I think another way is to state lesional min/max doses. Although, being a romanticist/nostalgist, I still like the "old lingo" of SRS. Either way, if I see a plan summary, I'd like to know lesional min/maxes, and an idea of the plan heterogeneity(ies).

@scarbrtj I respect the romanticism, but I think the old way is not helpful, and probably unnecessarily confusing for modulated plans. Lesion homogeneity has not correlated with control in any series of which I'm aware, and when you're treating >10 mets, you're not going to take the time to list out the dmax for every different lesion. We just report something like 20Gy/1fx to GTV 1-9, 16Gy/1fx to GTV 10 etc.

Also, I were treating a lesion in my own pons, I definitely would not make the dose more homogeneous, because that would increase the moderate isodose spill outside the tumor.

Here's an example from some data I haven't gotten around to writing up: 10 multi met plans, re-planned identically except that 20% hotspot was constrained on each target. Even without meeting the constraint, just the act of penalizing makes your plans crappier. Here's the 50% spill from one of the plans:

 

[evilbooyaa]

Didn't mean to out you inadvertently. I think what you guys are doing at UAB is cool as hell. I think the majority of the country can at least wrap their head around doing 2 out of the following 3 things:

1) Single plan/isocenter (for the entirety of the brain)
2) Single fraction
3) Multiple lesions

At my institution, we routinely do 2 of those 3 factors. Where we all kind of pause is doing all 3 of them simultaneously. Even like @nkmiami said, they do single isocenter, multiple lesion, but generally fractionate. Couple of our attendings do that here using TomoTherapy.

It's different when you have a couple of lesions clustered together in a small portion of the brain that you can put the iso within about 1cm of each of the lesions and not have to treat significant distance away. Again, my attendings are likely too anal or conservative with SRS as it is.

 

[emt409]

Didn't mean to out you inadvertently. I think what you guys are doing at UAB is cool as hell. I think the majority of the country can at least wrap their head around doing 2 out of the following 3 things:

1) Single plan/isocenter (for the entirety of the brain)
2) Single fraction
3) Multiple lesions

At my institution, we routinely do 2 of those 3 factors. Where we all kind of pause is doing all 3 of them simultaneously. Even like @nkmiami said, they do single isocenter, multiple lesion, but generally fractionate. Couple of our attendings do that here using TomoTherapy.

It's different when you have a couple of lesions clustered together in a small portion of the brain that you can put the iso within about 1cm of each of the lesions and not have to treat significant distance away. Again, my attendings are likely too anal or conservative with SRS as it is.

It's all good, my screenname was not exactly designed to be anonymous. I try not to say anything on the internet I wouldn't say in person or in an email anyway.

But if your attendings think that is controversial, I'd love to hear what they think about our coneless, frameless, MLC-based trigem and essential tremor thalamotomy treatments.

Here is a cool slide of the first guy we treated (off-trial b/c he also had brain mets) showing a nice dollop of 3-mo post-tx T1 enhancement where the treatment was.

 

[RadOncDoc21]

It's all good, my screenname was not exactly designed to be anonymous. I try not to say anything on the internet I wouldn't say in person or in an email anyway.

But if your attendings think that is controversial, I'd love to hear what they think about our coneless, frameless, MLC-based trigem and essential tremor thalamotomy treatments.

Here is a cool slide of the first guy we treated (off-trial b/c he also had brain mets) showing a nice dollop of 3-mo post-tx T1 enhancement where the treatment was.

View attachment 236329

Not giving away too much information but I went out to UAB and was impressed with how you guys are willing to push the envelope on so many levels. We just started doing frameless SRS to Trigems but nowhere near doing thalamotomy treatments but I'm slowly breaking my team down.

 

[evilbooyaa]

Lol the functional brain stuff is something that one of our attendings universally send out. I'm not sure if even I'm ever going to have the balls to give 100+ Gy single fraction for ET or Trigem without GK or at least a frame. Honestly I'll probably just refer those out even as an attending to a place that does them with frequency.

 

[medgator]

Lol the functional brain stuff is something that one of our attendings universally send out. I'm not sure if even I'm ever going to have the balls to give 100+ Gy single fraction for ET or Trigem without GK or at least a frame. Honestly I'll probably just refer those out even as an attending to a place that does them with frequency.

Completely agree. GK or bust for me

 

[emt409]

Lol the functional brain stuff is something that one of our attendings universally send out. I'm not sure if even I'm ever going to have the balls to give 100+ Gy single fraction for ET or Trigem without GK or at least a frame. Honestly I'll probably just refer those out even as an attending to a place that does them with frequency.

Completely agree. GK or bust for me

You think this takes balls? Just think about Lars Leksell was doing in the 60s. The first SRS was 200Gy WITH AN ORTHOVOLTAGE KV!!!!!! He used 250Gy I think for his initial bilateral anterior capsulotomies....craziness. Obviously orthovoltage was terrible, then he tried protons, then ended up developing gamma knife.

Anyway, Varian has licensed the IP of our MLC based functional radiosurgery technique, but hasn't integrated it yet. It's somewhere on the roadmap for a later version of HA I believe.

 

[evilbooyaa]

lol, I get where you're coming from Lars Leksell also didn't have to worry about getting sued. CYA medicine and swinging of the pendulum to at times worry excessively about toxicity (anecdotally) affects Rad Onc too. Regardless, you guys keep being like these guys:

 

[emt409]

LOL...this is one of my fave episodes of South Park.

 

[RadOncDoc21]

lol, I get where you're coming from Lars Leksell also didn't have to worry about getting sued. CYA medicine and swinging of the pendulum to at times worry excessively about toxicity (anecdotally) affects Rad Onc too. Regardless, you guys keep being like these guys:

View attachment 236336

Just think back when Timmerman was pushing SBRT in lung. Now it seems normal but I hear stories of how “crazy” it all seemed back then.

 

[scarbrtj]

Lesion homogeneity has not correlated with control in any series of which I'm aware

Someone posted something above re: MORE homogeneity, better control... https://www.researchgate.net/public...ain_Metastases_Implications_for_Local_Control
Of course, take w/ grain of salt. And, there have been some series in the AVM literature relating toxicities w/ IDLs/homogeneity/heterogeneity, "hot spots." Of course, back to semantics again, all these terms are mathematically interchangeable: homogeneity, heterogeneity, IDLs, "hot spot."

With fairly surprising consistency, for a well planned linac SRS treatment (1-9 mets in this series) V12 is roughly 2 x the treated volume + 2.5cc.

@emt409, let's consider just a solitary met, and per your paper "The prescription for all targets was standardized to 18 Gy" ...so with a 1 cc lesion, you would expect (and per your plot) a V12 of 4.5cc's. Assuming minimum lesional dose of 18 Gy, and that a typical plan for you would have ~150% "hot spot" (you said 130-170% somewhere), this would be akin to an Rx IDL of 67% ("normalizing" the 150% to 100%) and the V12 would be the 44% IDL. A 1cc lesion has about a 1.25 cm diameter; the 4.5 cc 44% IDL has a diameter of about 2.04 cm. This would mean the 44% IDL would be (2.04-1.25 divided by 2) 0.4 cm from the surface of the lesion (and the 50% IDL would be closer than that) implying a very rapid fall off from the 1cc target surface (a good plan). You can calc for other target cc's...

Target vol: 1cc, Distance of V12 from target surface: 0.40cm
Target volume: 2cc, Distance of V12 from target surface: 0.38cm
Target volume: 4cc, Distance of V12 from target surface: 0.37cm
Target volume: 8cc, Distance of V12 from target surface: 0.40cm
Target volume: 10cc, Distance of V12 from target surface: 0.41cm
Target volume: 15cc, Distance of V12 from target surface: 0.45cm
Target volume: 20cc, Distance of V12 from target surface: 0.48cm

Across target diameters of ~1-3 cm (target vol's of 1-20 cc), the V12 is changing linearly in relation to the target volume (obviously), but its distance from the target is sort of parabolic, sort of constant. What this implies to me is that in order to achieve keeping this falloff so remarkably consistent (~4-5 mm from target surface) no matter your lesion size, the planner is choosing a higher and higher "hot spot" as the lesion size goes up in size in order to make that happen. In other words, it seems the planner chooses (or allows for) more homogeneity for smaller lesions and less homogeneity for larger lesions. Of course this assumes equal minimal lesional doses across varying target volumes. And, if you're "choosing" less homogeneity for larger lesions to keep the V12 so tight, this means prescribing to lower IDLs for larger and larger lesions. Clearly well done plans on the larger lesions to keep that V12 so tight, but I suppose I choose lower (single shot) Rx doses for larger lesions versus keeping the Rx the same and prescribing to a lower and lower IDL as the lesion gets bigger and bigger. Or just hypofract the largers

 

[Palex80]

Do you guys use a Hexapod (or the Varian equivalent) to compensate for rotational shifts when doing single isocenter multiple brain mets? We don't have a Hexapod and that's keeping us away from treating with one isocenter.

 

[nkmiami]

Do you guys use a Hexapod (or the Varian equivalent) to compensate for rotational shifts when doing single isocenter multiple brain mets? We don't have a Hexapod and that's keeping us away from treating with one isocenter.

We dont presently have a 6 degree table, but use margin on the mets. Generally, I have pts come back the next day after simulation for a cone beam to check possible rotation extent, then decide with physics based on a formula what margin to add, where to place the isocenter, and will rarely use 2 isocenters.

It actually would be relatively cheap and easy to machine a head extender/adjuster to the couch that can rotate slightly (obviating the need for a table) and was told one university that did this (emory) some years ago. I was considering this but we just purchased vision rt, and they threw one of these devices/head adjuster in for us. Maybe someone smart can post a 3d printed solution.

 

[Gfunk6]

On the topic of multi-fraction SRS, do you guys use a surrogate (e.g. normal brain - GTV) to assess the risk of brain necrosis? I know for single-fraction the ideal V12 is < 5 cc (or < 10 cc max) but have not been able to locate a good one for 3 or 5 fractions.

 

[emt409]

Someone posted something above re: MORE homogeneity, better control... https://www.researchgate.net/public...ain_Metastases_Implications_for_Local_Control
Of course, take w/ grain of salt. And, there have been some series in the AVM literature relating toxicities w/ IDLs/homogeneity/heterogeneity, "hot spots." Of course, back to semantics again, all these terms are mathematically interchangeable: homogeneity, heterogeneity, IDLs, "hot spot."

@emt409, let's consider just a solitary met, and per your paper "The prescription for all targets was standardized to 18 Gy" ...so with a 1 cc lesion, you would expect (and per your plot) a V12 of 4.5cc's. Assuming minimum lesional dose of 18 Gy, and that a typical plan for you would have ~150% "hot spot" (you said 130-170% somewhere), this would be akin to an Rx IDL of 67% ("normalizing" the 150% to 100%) and the V12 would be the 44% IDL. A 1cc lesion has about a 1.25 cm diameter; the 4.5 cc 44% IDL has a diameter of about 2.04 cm. This would mean the 44% IDL would be (2.04-1.25 divided by 2) 0.4 cm from the surface of the lesion (and the 50% IDL would be closer than that) implying a very rapid fall off from the 1cc target surface (a good plan). You can calc for other target cc's...

Target vol: 1cc, Distance of V12 from target surface: 0.40cm
Target volume: 2cc, Distance of V12 from target surface: 0.38cm
Target volume: 4cc, Distance of V12 from target surface: 0.37cm
Target volume: 8cc, Distance of V12 from target surface: 0.40cm
Target volume: 10cc, Distance of V12 from target surface: 0.41cm
Target volume: 15cc, Distance of V12 from target surface: 0.45cm
Target volume: 20cc, Distance of V12 from target surface: 0.48cm

Across target diameters of ~1-3 cm (target vol's of 1-20 cc), the V12 is changing linearly in relation to the target volume (obviously), but its distance from the target is sort of parabolic, sort of constant. What this implies to me is that in order to achieve keeping this falloff so remarkably consistent (~4-5 mm from target surface) no matter your lesion size, the planner is choosing a higher and higher "hot spot" as the lesion size goes up in size in order to make that happen. In other words, it seems the planner chooses (or allows for) more homogeneity for smaller lesions and less homogeneity for larger lesions. Of course this assumes equal minimal lesional doses across varying target volumes. And, if you're "choosing" less homogeneity for larger lesions to keep the V12 so tight, this means prescribing to lower IDLs for larger and larger lesions. Clearly well done plans on the larger lesions to keep that V12 so tight, but I suppose I choose lower (single shot) Rx doses for larger lesions versus keeping the Rx the same and prescribing to a lower and lower IDL as the lesion gets bigger and bigger. Or just hypofract the largers

I think you need to do some single isocenter VMAT planning yourself (if you have Eclipse). The confusion between normalization practices is very common among people who trained in GK. You do not choose a hot spot in our technique. At all. Period. Via IDL selection, or optimization criteria. It is not in the cost function whatsover. The fluence patterns are much more complex in multi-met VMAT plans than GK or conformal arc plans, and you can't just infer relationships out of them like you are attempting to do.

Here is a link to our planning methodology: Dropbox - UAB RapidArc Stereotactic Radiosurgery Treatment Planning Guide - 10-15-15.pdf

As for the World Neurosurgery paper out of UVA, I hadn't seen it previously, and I need to read it in more detail; but I have some questions about the stats they report. Off the cuff, for PFS, their IDL 95% CI includes one, and also reports a p value of 0.02. This should not happen. Second, the IDL effect becomes larger (albeit still very small) on multivariate than univariate analysis. This indicates either a suppressor variable or sampling error. I can't think of any biological rationale why of their other variables would suppress the effect of an isodose line. Moreover, there is mathematically obligate covariance between IDL, marginal dose, and dmax, which doesn't appear to be accounted for in their model. Again, I need to read the paper in more detail, but I've sent papers I reviewed back for far less egregious statistical oddities.

 

[Palex80]

Does it choose you? Sorry. (The hotspot directly correlates to the minimum lesion dose/doses you choose... so it is chosen... perhaps it's better to say: one matters not if the spot is hot.)

Good point. I think the important clinical question is: Do we have data pointing out that a "too hot" hotspot within the GTV will increase the risk of radionecrosis and if so, what dose are we looking at. For the sake of the argument let's talk about an absolute and not a relative dose, since prescribing to different dose levels at the edge of the PTV makes it difficult to us to discuss this.

The question we should be asking is:

Is a X Gy hotspot (let's say for example 28 Gy) within the GTV a relevant factor for the development of clinical relevant radionecrosis?

 

[BobbyHeenan]

On the topic of multi-fraction SRS, do you guys use a surrogate (e.g. normal brain - GTV) to assess the risk of brain necrosis? I know for single-fraction the ideal V12 is < 5 cc (or < 10 cc max) but have not been able to locate a good one for 3 or 5 fractions.

Within the analysis of complications section of the retrospective one versus three fraction paper for larger brain mets, they suggested that (for the 27 Gy/3fx regimen) if the volume of normal brain getting 18 Gy was > 30cc they had a higher likelihood of radiation necrosis.

So that's one constraint I look at.

It's retrospective data with not a ton of events, so the quality of the data probably isn't great, but at least it's something.

If I can't meet that contstraint I typically drop down to 30 Gy in 5 fractions (rather than 27 in 3). Obviously, I don't have good data for this decision making, but it is what it is.

 

[BobbyHeenan]

Good point. I think the important clinical question is: Do we have data pointing out that a "too hot" hotspot within the GTV will increase the risk of radionecrosis and if so, what dose are we looking at. For the sake of the argument let's talk about an absolute and not a relative dose, since prescribing to different dose levels at the edge of the PTV makes it difficult to us to discuss this.

The question we should be asking is:

Is a X Gy hotspot (let's say for example 28 Gy) within the GTV a relevant factor for the development of clinical relevant radionecrosis?

I don't believe there is any robust data that suggests the hot spot within the GTV has much bearing on necrosis risk for met or AVM cases. Most series suggest it is the moderate isodose volumes (which obviously are often tied to target size) which predict for necrosis (ie 12 Gy volumes for single fraction).

Single fraction treatment of small mets on a gamma knife often can have hot spots of upwards of 40-44 Gy (not uncommon to see small mets get Rx'd 22 Gy to the 50% isodose line) and time and time again in publications (in the pre immunotherapy era at least) they show low chance of necrosis if a sub-cm met gets this kind of dose.

 

[scarbrtj]

I think you need to do some single isocenter VMAT planning yourself (if you have Eclipse)

AFAIK I was doing it and speaking on it antecedent you guys ...

You do not choose a hot spot in our technique. At all. Period. Via IDL selection, or optimization criteria. It is not in the cost function whatsover.

You do not directly choose your "hot spot" per se, I understand what you're saying and I'm not trying to be intentionally obtuse. However, due to your particular optimization (and structure creation) approach you are "choosing" increased "hot spot" levels than what other planners might in the course of their optimizations (perhaps in the way a person who chooses a healthy lifestyle "chooses" a longer lifespan).

Thanks for your Treatment Planning Guide. You guys turn some deeply ingrained definitions oppositely around! That is, what you call homogeneity and heterogeneity are actually reversed (in this instance from your instructions manual) versus the "universal definitions"... homogeneity is what you guys call equal minimum (and max?) doses between targets, intraplan, not the amount of "hot spot" within the target(s).

I did not see in the Treatment Planning Guide any optimization parameters for the NTO in Eclipse, and in this fashion the "hot spot" is in the cost function because as you push the NTO, the "hot spot" inside the tumor will go up. Do you use it or no? I find it's a pretty effective way to cut right to the fall-off issue, which is arguably the most important optimization/planning parameter (Paddick thought so too I guess). My idea is I want 50% of the absolute Gy Rx dose to be ~6-7mm (at most) from the target surface (I showed above where ~67% of the Rx dose is about 4-5 mm away from the target surface for you guys, so that probably roughly equates). For example, with NTO, I usually find myself using something like this:

Whereas something like this would be a much less rapid fall-off (and I wouldn't use it):

There are multiple ways to "skin the cat" in Eclipse, and many different roads lead to Rome. (For example, when I plan SRS in Eclipse and not looking at absolute dose isolines, I make Eclipse have the 100% IDL be the max relative IDL via the normalization GUI... "old school"... and it's exactly equivalent to NOT doing that... like @nkmiami says "philosophical" difference only.) IMHO, I've found the NTO method to be as good as the "control structure" (what we used to call "rings") method you mention, but quicker. Of course in the early days of planning, we didn't have the NTO option, and if NTO is not "what you were raised on," the control structure/"rings" method works great of course. (NTO is also like an indirect way to aim at a good Paddick GI.)

 

[scarbrtj]

Is a X Gy hotspot (let's say for example 28 Gy) within the GTV a relevant factor for the development of clinical relevant radionecrosis?

I'm gonna say, past a certain point, it has to be a relevant factor. Let me be sensationalist: within the known physics of our current machines and high energy X-rays as we deliver them, if my 1 cm spherical irradiation target in the brain had an interior "hot spot" of 80 Gy vs a "hot spot" of 20 Gy, there'd be more RN risk with the 80 Gy. Of course, you could not achieve equal minimal lesional doses (within the limits of physics) between these two "hot spot" parameters (the 80Gy target dose would be much higher because dose can only fall off so fast with X-rays and a linac), so in your mind's eye you can picture more high-dose spillage outside the target with the 80 Gy "hot spot" scenario and that's why there'd be more RN risk. From my experience, there's a point where you haven't pushed the "hot spot" enough, and a point where you've pushed it too much, in linac-based/Eclipse-based planning, and it kind of varies lesion-to-lesion. There's a "sweet spot." Looking at @emt409 paper/Treatment Planning Guide, e.g., I think it lives in the "sweet spot"... I am pretty sure not everyone planning SRS in Eclipse gets it quite so sweet.

 

[evilbooyaa]

Good point. I think the important clinical question is: Do we have data pointing out that a "too hot" hotspot within the GTV will increase the risk of radionecrosis and if so, what dose are we looking at. For the sake of the argument let's talk about an absolute and not a relative dose, since prescribing to different dose levels at the edge of the PTV makes it difficult to us to discuss this.

The question we should be asking is:

Is a X Gy hotspot (let's say for example 28 Gy) within the GTV a relevant factor for the development of clinical relevant radionecrosis?

Based off equivalent outcomes of GK and linac-based SRS for similar size and fractionation (which routinely prescribe to pretty wildly different isodose lines), I wouldn't overtly expect the bolded to be true for standard prescription doses (50% IDL for GK, 80% IDL for linac). Might be more of a concern for re-irradiation per the Wake Forest experience with re-irradiation to the same location (Using aggregate V40 as cut-off for RN incidence): Repeat stereotactic radiosurgery as salvage therapy for locally recurrent brain metastases previously treated with radiosurgery. - PubMed - NCBI

On the topic of multi-fraction SRS, do you guys use a surrogate (e.g. normal brain - GTV) to assess the risk of brain necrosis? I know for single-fraction the ideal V12 is < 5 cc (or < 10 cc max) but have not been able to locate a good one for 3 or 5 fractions.

Agree with V18 < 30cc as the cut-off for 3 fraction per Minitti. Although the paper focused on single metastases, I try to reach for this as possible even for multi-lesion plans using 3 fraction (or plan sum if using multiple isocenters). He does normal brain - GTV for his brain DVH evaluations as well.

 

[scarbrtj]

I wouldn't overtly expect the bolded to be true for standard prescription doses (50% IDL for GK, 80% IDL for linac).

It actually would be relevant, as @Palex80 phrased the question: "Is a X Gy hotspot (let's say for example 28 Gy) within the GTV a relevant factor for the development of clinical relevant radionecrosis?" and if you just varied the IDL Rx%:

The V12 would be much larger for higher IDL%.
EDIT: This is why, early in our thread, the method of just reporting minimum lesional dose, with no info about plan Dmax or the "hot spots" with the targets, leaves me more uncertain than I'm used to re: the dose fall-off from the lesion surface. With Rx IDL% and a minimum lesional dose I have a better mental gestalt about that. Over a certain range w/ the linac and Eclipse, the V12 is growing/decreasing in size in relation to target (if the target is getting a min 18 Gy e.g.) depending on how you "choose" your hotspot, aka how you "choose" your homogeneity, aka how you "choose" what the target DVH looks like, aka how you normalize dose to your target, etc etc.

 

[scarbrtj]

@emt409 in the Treatment Planning Guide it says:

3.8.2. For each other target, adjust the priority of its optimization constraint such that its DVH curve is covering a dose the same difference from the anchor curve as the difference between the two targets’ respective prescription doses.
3.8.2.1. For example, prescription dose for target A is 18 Gy and is 24 Gy for target B (Δ = 6 Gy). Target A is designated as the anchor target. If during the initial phase of the optimization, Target A’s DVH curve indicates 100% coverage at 12 Gy, then adjust target B’s optimization constraint priority such that its curve indicates 100% coverage at 12 + 6 or 18Gy. Thus, when the post-optimization plan-wide normalization is performed, each target should have roughly 100% coverage at its respective prescription dose.

Should it not be:
3.8.2. For each other target, adjust the priority of its optimization constraint such that its DVH curve is covering a dose the same relative difference from the anchor curve as the relative difference between the two targets’ respective prescription doses.
3.8.2.1. For example, prescription dose for target A is 18 Gy and is 24 Gy for target B (Δ = +33%). Target A is designated as the anchor target. If during the initial phase of the optimization, Target A’s DVH curve indicates 100% coverage at 12 Gy, then adjust target B’s optimization constraint priority such that its curve indicates 100% coverage at 12 + 33% or 16 Gy. Thus, when the post-optimization plan-wide normalization is performed to achieve roughly 100% 18 Gy coverage of target A, each target should have roughly 100% coverage at its respective prescription dose.

Because if you have A @12Gy and B@16 Gy from optimization, normalization affects everything equally and if you up-normalize A to 18 Gy then B will go to 24 Gy w/normalization... as it will always be 33% relatively higher... it won't always be 6 Gy absolutely higher as you normalize after optimization, yes? Else if you optimize A to 12Gy and B to 18Gy, then when you normalize A to 18 Gy, B will go to 27 Gy.

 

[emt409]

@emt409 in the Treatment Planning Guide it says:

3.8.2. For each other target, adjust the priority of its optimization constraint such that its DVH curve is covering a dose the same difference from the anchor curve as the difference between the two targets’ respective prescription doses.
3.8.2.1. For example, prescription dose for target A is 18 Gy and is 24 Gy for target B (Δ = 6 Gy). Target A is designated as the anchor target. If during the initial phase of the optimization, Target A’s DVH curve indicates 100% coverage at 12 Gy, then adjust target B’s optimization constraint priority such that its curve indicates 100% coverage at 12 + 6 or 18Gy. Thus, when the post-optimization plan-wide normalization is performed, each target should have roughly 100% coverage at its respective prescription dose.

Should it not be:
3.8.2. For each other target, adjust the priority of its optimization constraint such that its DVH curve is covering a dose the same relative difference from the anchor curve as the relative difference between the two targets’ respective prescription doses.
3.8.2.1. For example, prescription dose for target A is 18 Gy and is 24 Gy for target B (Δ = +33%). Target A is designated as the anchor target. If during the initial phase of the optimization, Target A’s DVH curve indicates 100% coverage at 12 Gy, then adjust target B’s optimization constraint priority such that its curve indicates 100% coverage at 12 + 33% or 16 Gy. Thus, when the post-optimization plan-wide normalization is performed to achieve roughly 100% 18 Gy coverage of target A, each target should have roughly 100% coverage at its respective prescription dose.

Because if you have A @12Gy and B@16 Gy from optimization, normalization affects everything equally and if you up-normalize A to 18 Gy then B will go to 24 Gy w/normalization... as it will always be 33% relatively higher... it won't always be 6 Gy absolutely higher as you normalize after optimization, yes? Else if you optimize A to 12Gy and B to 18Gy, then when you normalize A to 18 Gy, B will go to 27 Gy.

You are very astute, and indeed correct, I inadvertently sent you the 2015 recipe. We didn't noticed that error until we did a comparative study on normalizations. It should be relative. It's fixed in the 2016 version.

AFAIK I was doing it and speaking on it antecedent you guys ...

You do not directly choose your "hot spot" per se, I understand what you're saying and I'm not trying to be intentionally obtuse. However, due to your particular optimization (and structure creation) approach you are "choosing" increased "hot spot" levels than what other planners might in the course of their optimizations (perhaps in the way a person who chooses a healthy lifestyle "chooses" a longer lifespan).

Thanks for your Treatment Planning Guide. You guys turn some deeply ingrained definitions oppositely around! That is, what you call homogeneity and heterogeneity are actually reversed (in this instance from your instructions manual) versus the "universal definitions"... homogeneity is what you guys call equal minimum (and max?) doses between targets, intraplan, not the amount of "hot spot" within the target(s).

I did not see in the Treatment Planning Guide any optimization parameters for the NTO in Eclipse, and in this fashion the "hot spot" is in the cost function because as you push the NTO, the "hot spot" inside the tumor will go up. Do you use it or no? I find it's a pretty effective way to cut right to the fall-off issue, which is arguably the most important optimization/planning parameter (Paddick thought so too I guess). My idea is I want 50% of the absolute Gy Rx dose to be ~6-7mm (at most) from the target surface (I showed above where ~67% of the Rx dose is about 4-5 mm away from the target surface for you guys, so that probably roughly equates). For example, with NTO, I usually find myself using something like this:

Whereas something like this would be a much less rapid fall-off (and I wouldn't use it):

There are multiple ways to "skin the cat" in Eclipse, and many different roads lead to Rome. (For example, when I plan SRS in Eclipse and not looking at absolute dose isolines, I make Eclipse have the 100% IDL be the max relative IDL via the normalization GUI... "old school"... and it's exactly equivalent to NOT doing that... like @nkmiami says "philosophical" difference only.) IMHO, I've found the NTO method to be as good as the "control structure" (what we used to call "rings") method you mention, but quicker. Of course in the early days of planning, we didn't have the NTO option, and if NTO is not "what you were raised on," the control structure/"rings" method works great of course. (NTO is also like an indirect way to aim at a good Paddick GI.)

Considering we were one of the first centers in the world to deploy RapidArc, and one of my colleagues published the feasibility paper, I'd be very impressed if you were doing this antecedent to us

The NTO works well for one, two, and sometimes even three mets, depending on how far apart they are. As you are currently utilizing the NTO, you are simultaneous enforcing a ceiling and a floor constraint on any targets within 10cm of each other.

The reason the SRS NTO (e.g. the secret sauce in HA) was developed was because the NTO didn't cut it for higher numbers of targets. Also, the PO algorithm (which it appears you have selected based on the optimizer interface) integrates even the regular NTO better than the PRO did in both v13 and v15 of Eclipse.

 

[nkmiami]

I have no experience with brain lab or adac, but both have similar techniques to this. I saw a demonstration of the brainlab, where it did a very good job of autocontouring the lesions on mri and critical structures and then autoplanning. I imagine that is the future of this technology.

 

[evilbooyaa]

I have no experience with brain lab or adac, but both have similar techniques to this. I saw a demonstration of the brainlab, where it did a very good job of autocontouring the lesions on mri and critical structures and then autoplanning. I imagine that is the future of this technology.

Unfortunately contouring bright little dots on an MRI isn't all that hard. Not entirely sure why we need auto-contouring for that, but I guess some people really are that lazy. It's likely going to mess up on vascular structures near the lesion (including or excluding things that shouldn't be) as well as potentially if tumor interfaces with bone.

 

[scarbrtj]

I'd be very impressed if you were doing this antecedent to us

Heh. It's a close call temporally. Aug 2007 slide here. Had just started RapidArc (although RapidArc isn't/wasn't necessary for single iso SRS; we had done static beam IMRT for these). I know UAB was very early w/ it too; #1 or #2, IDK. (As I recall, first RapidArc-ing anywhere was 2007.) First here of course doesn't necessarily mean smart or progressive; it really means good graces w/ Varian coupled with buying new machines from them in the mid-2000s

 

[nkmiami]

Unfortunately contouring bright little dots on an MRI isn't all that hard. Not entirely sure why we need auto-contouring for that, but I guess some people really are that lazy. It's likely going to mess up on vascular structures near the lesion (including or excluding things that shouldn't be) as well as potentially if tumor interfaces with bone.

well if you are treating 20 mets.. the idea is to make this as easy whole brain- It is hippocampal and to some extent hair sparing-UAB recently describe a simulation where you would need about 40 mets before cumulative mean whole brain dose is 8 gy- maybe emt can comment on this.

Radiosurgery of brain metastases — How many is too many?

 

[scarbrtj]

The NTO works well for one, two, and sometimes even three mets, depending on how far apart they are.

For me w/ my own patient selection and private practice proclivities, 1-2/100 single-iso SRS patients will have 3 mets, ~10/100 2 mets, and ~zero out of 100 will have 4 or more. Once it gets up into 4, or more, especially if they're spread out, I do two iso's. I now know why: 4 was tough to plan satisfactorily, ha.

 

[scarbrtj]

UAB recently did a simulation where you would need about 40 mets before cumulative mean whole brain dose is 8 gy

The time between sim and SRS treat for a patient with 40 brain mets needs to be REALLY quick.

 

[evilbooyaa]

well if you are treating 20 mets.. the idea is to make this as easy whole brain- It is hippocampal and to some extent hair sparing-UAB recently describe a simulation where you would need about 40 mets before cumulative mean whole brain dose is 8 gy- maybe emt can comment on this.

Radiosurgery of brain metastases — How many is too many?

Just because we can do it doesn't mean that we should. Anything is feasible, but whether it should be part of clinical practice I think is the more important question.

 

[nkmiami]

Just because we can do it doesn't mean that we should. Anything is feasible, but whether it should be part of clinical practice I think is the more important question.

I am not advocating it, but I could see where software takes this to a point where it is easier and more convenient than whole brain for the pt without requiring any more resources/planning time. It happened with IMRT in many sites...

 

[medgator]

I am not advocating it, but I could see where software takes this to a point where it is easier and more convenient than whole brain for the pt without requiring any more resources/planning time. It happened with IMRT in many sites...

Sure, but I still don't see table time being faster than two opposed laterals... no matter how fast that dose rate gets

 

[nkmiami]

Sure, but I still don't see table time being faster than two opposed laterals... no matter how fast that dose rate gets

Our dosimetrist will mix in some high energy and field within a field to keep it fresh.

 

[emt409]

Just because we can do it doesn't mean that we should. Anything is feasible, but whether it should be part of clinical practice I think is the more important question.

Yea, tbh, the point of that paper was more of a dosimetric commentary. I.e. You should not need to base your decision to WBRT on the dosimetry of high number of mets because the quality of linac SRS has improved so significantly. Now, you really need to invest thought into the clinical decision to treat large numbers met. And it's becoming an increasingly complicated decision, in need of much more evidence to support practice guidelines.

Factors to take into account: performance status, life expectancy, histology, systemic disease burden, patient baseline cognitive ability, likelihood of occult microscopic disease, patient likelihood of follow-up compliance, concurrence/timing of systemic therapy.

I actually dx'd MEN1 in guy inpatient yesterday, and he has probably 20 pancreatic neuroendocrine mets in the brain. Even though he is completely asymptomatic, has KPS 100, and we have treated higher number of mets both single and 5 fx tx, we are going to WBRT him because the likelihood of occult disease is so high, and the likelihood of achieving systemic disease control is so low.

On the other hand, we have a lot of patients from Huntsville, where I grew up, that are NASA engineers, or have other highly technical jobs, and deeply value their cognitive ability. In several cases I can think of, they perhaps from a clinical decision making perspective might really have been best served by WBRT, but refused treatment unless it was SRS. So SRS they got.

 

[medgator]

I'm curious to what your experience has been with this in the community. For brain mets 1.5 cm and less not near critical structures, I always treat single fraction. Usually 18-20 Gy x1 Rx to the 70-90% IDL. The only time I fractionate is if the size is > 1.5 cm or if it is near critical structure.

I understand in the community you sometimes have to compromise based on your hardware availability (e.g. no single fraction SRS if you have 1 cm MLCs). However, I've seen people with state of the art accelerators (e.g. TrueBeam/VersaHD) who treat 5-6 Gy x 5 or 8-9 Gy x 3. These are for unresected brain mets not resection cavity s/p surgery.

Less BED = inferior LC rates. The cynic in me says they do it for money, but curious what others think.

https://www.redjournal.org/article/S0360-3016(18)33927-0/pdf

 

[scarbrtj]

For the record: BED10 of 27/3 and 18/1 are equivalent.

Yes. Wasn't it Hall (Withers?) who said something like "Treated well in one fraction is treated better in three"?

 

[evilbooyaa]

https://www.redjournal.org/article/S0360-3016(18)33927-0/pdf

Reaffirmation of the Minniti Data.

Focus continues to be on anything greater than 2cm. I think 1-year LC rates of 78% vs 93% is significant, even if the P-value on the meta-analysis isn't. Obviously the main draw is for decreasing incidence of radionecrosis. The paper's not perfect - how did the smaller tumors have a higher RN rate with SF than larger tumors? Due to changes in single fraction dosing?

 

[wandering star]

you can run a mathematical sim of the old saying "daily setup error is plus or minus a half cm" and show that PTV margins need to be in the ~1.4-1.5 cm range (which equals the old daily block margins of 2 cm with just daily skin mark setup). Daily bony anatomy setup would get you in the PTV=1 range, possibly. But this is true daily setup error just to skin marks. Every maneuver--daily port films, CBCT, framed stereotaxy--makes it smaller. The setup errors with or without IGRT are always normally distributed unidimensionally yet behave in an "anti-magnetic" fashion about the isocenter. I tried to coin a term "spheroprobability"... it never caught on.

This is a very late reply but you may be interested in knowing that you have rediscovered the chi-squared distribution with 3 degrees of freedom. Technically you are using the square root of it, which is simply the chi distribution. The "3" in 3 degrees of freedom refers to the number of dimensions (X, Y, and Z). This is a surprising and unusual connection between radiation physics, the number of dimensions of the world we live in, and probability/statistics.

If you look at the wikipedia page for the chi distribution, you find its peak (with 3 degrees of freedom) is at about 1.4. Considering that you have 5 mm std dev in x/y/z, you would predict that the peak error is 7 mm, which exactly matches your infographic.

 

[taserlaser]

This is a very late reply but you may be interested in knowing that you have rediscovered the chi-squared distribution with 3 degrees of freedom. Technically you are using the square root of it, which is simply the chi distribution. The "3" in 3 degrees of freedom refers to the number of dimensions (X, Y, and Z). This is a surprising and unusual connection between radiation physics, the number of dimensions of the world we live in, and probability/statistics.

If you look at the wikipedia page for the chi distribution, you find its peak (with 3 degrees of freedom) is at about 1.4. Considering that you have 5 mm std dev in x/y/z, you would predict that the peak error is 7 mm, which exactly matches your infographic.

I’m just going to go out on a limb and do some congratulations here, and say damn I love this field and the people we interact with. For all the gloom and doom that’s been around, the day to day patient stuff but also the math/physics/nerdy stuff really puts this job in such a great little niche.

 

[scarbrtj]

This is a very late reply but you may be interested in knowing that you have rediscovered the chi-squared distribution with 3 degrees of freedom. Technically you are using the square root of it, which is simply the chi distribution. The "3" in 3 degrees of freedom refers to the number of dimensions (X, Y, and Z). This is a surprising and unusual connection between radiation physics, the number of dimensions of the world we live in, and probability/statistics.

If you look at the wikipedia page for the chi distribution, you find its peak (with 3 degrees of freedom) is at about 1.4. Considering that you have 5 mm std dev in x/y/z, you would predict that the peak error is 7 mm, which exactly matches your infographic.

Thanks for pointing this out. At the time I did feel like I was probably treading over some previously discovered/fundamental ground but I didn't have the math chops/background to describe what I was seeing. Some time later I would come across a Rayleigh distribution and I thought that was close; turns out it's chi with n=2 freedom. The n=3 freedom better matches the 3D curve. (Slow vs immediate uptick from zero.) The crux of my observation (that patients tend to set up at the isocenter vs patients almost never set up at the isocenter, the former being false and the latter true) has not been fully explored in rad onc (mathematically or clinical implications-wise). Of course, I would say that. Further Wikipedia-ing shows me that eponymously the distribution is a Maxwell-Boltzmann distribution. Pretty cool. Einstein wouldn't have been Einstein without Maxwell. Maybe I can be somebody's Maxwell one day... who knows!

 

[wandering star]

The crux of my observation (that patients tend to set up at the isocenter vs patients almost never set up at the isocenter, the former being false and the latter true) has not been fully explored in rad onc (mathematically or clinical implications-wise).

One way of saying this is that you're playing darts and you're trying to nail the bullseye. If you are a darts player with sure aim and you try to hit the bullseye at the center, you're still more likely to hit the circle around it, not because of some systematic error but simply because the circle around it has a lot more area. In the same way, even though we try to center patients correctly, we're more likely to miss by a few mm simply because that zone has much more volume.

 

[scarbrtj]

One way of saying this is that you're playing darts and you're trying to nail the bullseye. If you are a darts player with sure aim and you try to hit the bullseye at the center, you're still more likely to hit the circle around it, not because of some systematic error but simply because the circle around it has a lot more area. In the same way, even though we try to center patients correctly, we're more likely to miss by a few mm simply because that zone has much more volume.

The same reason why the question "Which part of the spine is prostate cancer most likely to spread: cervical, thoracic, or lumbar region?" has the answer "thoracic." Why? Because there are more thoracic vertebral bodies (and spine volume in the thoracic bodies) than cervical or lumbar.

 

[Lawpy]

I’m just going to go out on a limb and do some congratulations here, and say damn I love this field and the people we interact with. For all the gloom and doom that’s been around, the day to day patient stuff but also the math/physics/nerdy stuff really puts this job in such a great little niche.

This is definitely a good thread since I didn't expect to see stuff like chi distributions posted on SDN

 

[Palex80]

The same reason why the question "Which part of the spine is prostate cancer most likely to spread: cervical, thoracic, or lumbar region?" has the answer "thoracic." Why? Because there are more thoracic vertebral bodies (and spine volume in the thoracic bodies) than cervical or lumbar.

Is there any data for that actually? I seem to have the impression that prostate cancer cells tend to metastasize in the lumbar spine and pelvis more often than for example in the cervical spine or the skull bones...

 

[scarbrtj]

Is there any data for that actually? I seem to have the impression that prostate cancer cells tend to metastasize in the lumbar spine and pelvis more often than for example in the cervical spine or the skull bones...

If you have a multiple choice "Where does prostate cancer more commonly metastasize? A. C-spine. B. T-spine. C. L-spine" as phrased I think the choice is T-spine. If you add in pelvis, etc., IDK.