Fidelity of deformed MRI fusion

[Gfunk6]

Our practice uses MIM for contouring and its deformation algorithm for fused MRIs. I am curious to how much practicing RO's "trust" the fusion.

Obviously, getting a new MRI in the treatment position is the ideal solution but is not practical for all of us. Let's say you have a patient with a brain or spine treatment site and are fusing either with the planning CT. It is clear to you that rigid fusion doesn't work because of changes in positioning. What do you guys do?

1. Try to do a box-based rigid fusion in the area of interest?
2. Try to do a complete MRI deformation?

On the occasions I've done #2 in MIM, the results look very good but the question is how much this represents reality for you to utilize it to contour OARs.

---

Members don't see this ad.

 

[mheat3]

Our practice uses MIM for contouring and it's deformation algorithm for fused MRIs. I am curious to how much practicing RO's "trust" the fusion.

Obviously, getting a new MRI in the treatment position is the ideal solution but is not practical for all of us. Let's say you have a patient with a brain or spine treatment site and are fusing either with the planning CT. It is clear to you that rigid fusion doesn't work because of changes in positioning. What do you guys do?

1. Try to do a box-based rigid fusion in the area of interest?
2. Try to do a complete MRI deformation?

On the occasions I've done #2 in MIM, the results look very good but the question is how much this represents reality for you to utilize it to contour OARs.

For brain I always try to do a rigid fusion and selectively cone down the ROI until I see anatomical landmarks (8th nerve, etc) are lined up. Usually you can get the MRI rigid correct with a relatively large ROI.

Spine is a whole other ballgame. The issue and concern for me is that cord can move ant to post or side to side depending on how the patient was laying on the table. If I don't think the MRI spine was done in treatment planning position I tend to be more cautious with my OAR PRV as well as what does willing to accept for spine SBRT.

In general I am not a big fan of deformable registration. Something never feels quite right (but that is purely based on my gut feel, no data I think).

Curious to hear others thoughts as well.

 

[evilbooyaa]

I really dislike deformable registration. As above, just a 'gut feeling' of not trusting it. If I was going to use it for fractionated RT I'd probably give a PRV to whatever OARs you've drawn for uncertainty that can only be identified on MRI. I really dislike using a deformable registered MRI for GTV contouring (say a NasoPhx/Sinonasal primary or something).

For brain SRS I would do option 1 without a doubt in contouring GTVs.

For Spine SRS would give larger PRV to cord.

 

[OTN]

I use deformable MRI registration with liver SBRT where the tumor is tough to see on CT. Valuable there, as there's not much else you can do. I simulate with belt-based vs plate-based abdominal compression, which seems to help the fusion.

 

[RickyScott]

Our practice uses MIM for contouring and it's deformation algorithm for fused MRIs. I am curious to how much practicing RO's "trust" the fusion.

Obviously, getting a new MRI in the treatment position is the ideal solution but is not practical for all of us. Let's say you have a patient with a brain or spine treatment site and are fusing either with the planning CT. It is clear to you that rigid fusion doesn't work because of changes in positioning. What do you guys do?

1. Try to do a box-based rigid fusion in the area of interest?
2. Try to do a complete MRI deformation?

On the occasions I've done #2 in MIM, the results look very good but the question is how much this represents reality for you to utilize it to contour OARs.

Also have mim and Been dealing with this for years and have employed both 1 and 2 and sometimes used composite volumes. Sometimes mris have spatial distortion, and manufacturers detail sequences to reduce it, but often Insurance does not let you getting 2nd mri. Find that ct sim with iv contrast is also very helpful.
Why I am a big fan of putting a margin and giving several 3-5 fractions.

 

[radiation]

Most of the time a rigid fusion for the brain should be pretty close unless some anatomy change has happened between mri and ct (ie surgery). MIM has a nice "sanity check" for deformable registration where it overlays a grid that shows you exactly what was deformed. You can really go down the rabbit hole with their advanced features that let you do a mix of rigid registrations for important landmark structures like brainstem and deformable parts like cavities.

Brainlab also has a version of this in the spine that automatically keeps rigid registration of vertebral bodies, but allows for deformable registration to account for flexion/extension changes

Spare the Spinal Cordwith Steep Dose Falloff - Brainlab

The fully automated workflow is optimized for the treatment of spinal metastases and produces very steep dose falloff within the affected vertebrae.

www.brainlab.com

 

[Gfunk6]

I would also agree for brain that we use rigid fusion 99% of the time. However, there are certainly cases where we do SRS to a resection cavity where the shape and configuration of the MRI and planning CT are not precisely the same. In those cases I feel safe using deformable registration as I have the CT images to verify anatomy.

I also agree spine and other sites is complicated. Other than "gut" is there any good quantitative data people are aware of to trust deformable MR registration?

 

[Neuronix]

Always rigid. I'd never base my contours for SRS on a deformable brain registration.

I don't think MIM does a very good job with brain fusions compared to Brainlab or Gamma Plan. Physics or I personally have to mess around with MIM for awhile in many cases, and sometimes I still think my fusion is off by a tiny amount despite repeated attempts--though I am very picky.

There should never be a need for deformation inside the skull. Head position doesn't matter, because the skull is a rigid structure and brain within that skull will essentially always be in the same place unless there is a time difference between the CT and MRI. I try to get MRI and CT on same day or very close in time for this reason.

The things that deform are the spine/neck, eye position/optic nerve position, tongue, etc. So make sure the fusion isn't using any information from those areas. Especially 3T MRI at the surface or around sinuses can have some small geometric distortions, so what I look at most for fusions is the tentorium and falx. These complex and large structures look very similar on CT and MRI and are centrally located.

Sometimes I get a CT w/ contrast or do a CT w/ contrast sim if I really can't get a new MRI for some reason and have to rely on CT. CT always looks a bit different than MRI, so I try to use a fused MRI every time for consistency.

Spine is a whole different story. I do MRI simulations to get the patient in the same position.

 

[scarbrtj]

Our practice uses MIM for contouring and its deformation algorithm for fused MRIs. I am curious to how much practicing RO's "trust" the fusion.

Obviously, getting a new MRI in the treatment position is the ideal solution but is not practical for all of us. Let's say you have a patient with a brain or spine treatment site and are fusing either with the planning CT. It is clear to you that rigid fusion doesn't work because of changes in positioning. What do you guys do?

1. Try to do a box-based rigid fusion in the area of interest?
2. Try to do a complete MRI deformation?

On the occasions I've done #2 in MIM, the results look very good but the question is how much this represents reality for you to utilize it to contour OARs.

In these situations, I diversify the uncertainty portfolio and create something like a GTVr (rigid) and GTVd (deform) from two separate MRI matching process for a hybridized GTVrd. And then put a 1mm/tight margin around that. Knowing that it is highly likely the target of interest for me is covered. It creates a slightly larger GTV yes, but it appeals to me more than a uniform PTV expansion to admit and deal with the uncertainty. No matter what data comes out in the future, I think it’s very unlikely this approach would show suboptimal LC.

 

[RickyScott]

Always rigid. I'd never base my contours for SRS on a deformable brain registration.

I don't think MIM does a very good job with brain fusions compared to Brainlab or Gamma Plan. Physics or I personally have to mess around with MIM for awhile in many cases, and sometimes I still think my fusion is off by a tiny amount despite repeated attempts--though I am very picky.

There should never be a need for deformation inside the skull. Head position doesn't matter, because the skull is a rigid structure and brain within that skull will essentially always be in the same place unless there is a time difference between the CT and MRI. I try to get MRI and CT on same day or very close in time for this reason.

The things that deform are the spine/neck, eye position/optic nerve position, tongue, etc. So make sure the fusion isn't using any information from those areas. Especially 3T MRI at the surface or around sinuses can have some small geometric distortions, so what I look at most for fusions is the tentorium and falx. These complex and large structures look very similar on CT and MRI and are centrally located.

Sometimes I get a CT w/ contrast or do a CT w/ contrast sim if I really can't get a new MRI for some reason and have to rely on CT. CT always looks a bit different than MRI, so I try to use a fused MRI every time for consistency.

Spine is a whole different story. I do MRI simulations to get the patient in the same position.

Do you ever see enhancement on CT w/contrast that is outside MRI defined gtv?

 

[Neuronix]

I think it’s very unlikely this approach would show suboptimal LC.

The bigger risk IMO is regrowth of tumor from using old MRIs. There is one paper showing that using 14 day old MRIs or older results in inferior control. There is another paper showing significant changes in 7+ day old MRIs requiring re-contouring compared to MRI on day of SRS.

I treat a lot of brain mets and glioblastomas, and a lot can change in 2 weeks. I mean sure most of the time not much changes, but sometimes there can be significant enhancement adjacent to cavity that wasn't there post-op or some satellite or multi-focal nodule that pops up in the interval.

Do you ever see enhancement on CT w/contrast that is outside MRI defined gtv?

Good question. The CT w/ contrast usually looks smaller than the MRI post-gad in my opinion. I don't do this often though, so this is based on a handful of cases.

 

[scarbrtj]

The bigger risk IMO is regrowth of tumor from using old MRIs. There is one paper showing that using 14 day old MRIs or older results in inferior control. There is another paper showing significant changes in 7+ day old MRIs requiring re-contouring compared to MRI on day of SRS.

I treat a lot of brain mets and glioblastomas, and a lot can change in 2 weeks. I mean sure most of the time not much changes, but sometimes there can be significant enhancement adjacent to cavity that wasn't there post-op or some satellite or multi-focal nodule that pops up in the interval.

What I’m saying is recent mri. As recent as possible. By two separate processes I mean virtual MRI matching processes. Getting MRI on same day as tx? Deform it. Get target volume 1. “Paste” on CT. Rigid match it. Get TV2. “Paste” on CT. Boolean [AND] 1 and 2. It’s like saying “Which is right: rigid or deform? both.”

 

[Neuronix]

What I’m saying is recent mri. As recent as possible. By two separate processes I mean virtual MRI matching processes.

Fair enough. We're in agreement on that.

What I meant is that the most common error I see and the reason why people usually struggle with CT-MRI matching is because the MRI is too old. If the MRI and CT are done on same day or very close in time, I don't see why you would ever need the deformable registration.

I've done hundreds of intracranial SRS cases and never once deformed or wanted to deform. But I'm very open to there being something I'm missing.

I guess since I'm thinking about it, if the CT and MRI are very close to the time of surgery there can be a lot of change. Spacing things out a bit from surgery (I aim for treatment ~4 weeks from surgery) also means that usually the cavity has stabilized by the time you're doing pre-treatment imaging.

 

[scarbrtj]

Fair enough. We're in agreement on that.

What I meant is that the most common error I see and the reason why people usually struggle with CT-MRI matching is because the MRI is too old. If the MRI and CT are done on same day or very close in time, I don't see why you would ever need the deformable registration.

I've done hundreds of intracranial SRS cases and never once deformed or wanted to deform. But I'm very open to there being something I'm missing.

I guess since I'm thinking about it, if the CT and MRI are very close to the time of surgery there can be a lot of change. Spacing things out a bit from surgery (I aim for treatment ~4 weeks from surgery) also means that usually the cavity has stabilized by the time you're doing pre-treatment imaging.

Correct. I’m big on setting myself up for treatment planning success too.

 

[radiation]

I also agree spine and other sites is complicated. Other than "gut" is there any good quantitative data people are aware of to trust deformable MR registration?

https://s3.amazonaws.com/downloads.mimsoftware.com/Deformable_Image_Registration_in_MIM_Maestro_Evaluation_and_Description.pdf

MIM does allow you to quantify the deformation. The quality of the deformation is dependent on the case, so it will be really hard to find a useful study on this. When I have used, I will overlay the deformable grid and see if the areas that were deformed are acceptable. You can get the actual numerical values of the deformation translations, but not sure if that sort of quantitative data is useful

 

[radiation]

Fair enough. We're in agreement on that.

What I meant is that the most common error I see and the reason why people usually struggle with CT-MRI matching is because the MRI is too old. If the MRI and CT are done on same day or very close in time, I don't see why you would ever need the deformable registration.

I've done hundreds of intracranial SRS cases and never once deformed or wanted to deform. But I'm very open to there being something I'm missing.

I guess since I'm thinking about it, if the CT and MRI are very close to the time of surgery there can be a lot of change. Spacing things out a bit from surgery (I aim for treatment ~4 weeks from surgery) also means that usually the cavity has stabilized by the time you're doing pre-treatment imaging.

I agree with this. One interesting thing I have found is that gamma knife people almost always do same day MRI and treat because of how the workflow was developed with frame so fusions are almost never an issue. People that use Cyberknife or Linac almost always have a gap between MRI, sim, and treatment and seem to have more fusion issues

 

[communitydoc13]

MRI basics for radiation oncologists This is a great article on MRI use for radoncs. I always use rigid registration and other than insisting on 3D , thin slice image sets, I don't specify special image series for treatment planning. The intrinsic geometric fidelity of the MRI is interesting to think about. There are intrinsic deformities to the MRI modality (order of magnitude 1-2 mm) and I wonder if deformable registration may actually be better for small fields as its distorting the already distorted MRI to the much more geometrically faithful CT scan. Everything I do is linac based, and I pretty much always use 2mm margins.

 

[seper]

Agree, rigid fusion is the way to go, specifically for brain MRI