Tractography in the presence of Edema

1

Dear MRTrix community,

Could you please clarify me on how the tractography is being performed in the presence of Edema? Is there any default correction that is being applied? I couldn’t find my answer in the documentation. So I’d appreciate any guidance on this matter.

Thank you,
Parand

1 Like
2

Hi Parand,

There’s certainly not any mechanism that decides “this voxel contains edema; let’s perform tractography differently, and not tell the user about it”. That would be very much against our ethos.

This question necessitates answering in two parts:

  1. How is deconvolution affected?

    One must consider first how the DWI signal is changed by the presence of edema, and then correspondingly how the parameters of a fitted model may be changed by such. If performing multi-tissue deconvolution, then one would expect that hyperintensity in the b=0 data will be primarily attributed to the CSF-like compartment. But there could also be changes in T2, which, due to the way in which we deal with intensity normalisation, could influence the magnitude of the WM ODF, and that in turn could have an influence on tractography.

  2. How is streamlines tractography affected?

    1. Once you get to the point of performing tractography, it’s only the WM ODFs that are utilised; the rest of the decomposition is never touched. So the question is how may the WM ODFs differ in a way that would be consequential for tractography. This could be a reduction in overall amplitude, which if severe enough may result in completely forbidding streamlines from crossing due to failing to exceed the FOD amplitude criterion. Or it could be an increase in overall amplitude if the T2 of intra-cellular water is affected or the extra-cellular water becomes more restricted. Exactly what happens here depends on the nuances of how any given streamlines tractography algorithm operates.

    2. Note that you should certainly not expect a correspondence between the magnitudes of WM ODFs in pathology and the density of streamlines. While the SIFT(2) methods attempt to establish correspondence between these two parameters, they (and related methods) do so on the premise of continuity of fibre density along each trajectory. If a localised pathology such as edema violates this premise (whether due to artifacts of the diffusion model or genuine biological discontinuity), this correspondence can’t be established fully, so these methods can’t wholly “correct” the recosntruction, and the metric of connectivity gets biased (just as happens for other metrics).
      Here’s a relevant discussion if you feel like consuming some Cancer (thankfully there’s mutual understanding and a common goal now).

    3. The behaviour of tractography may alternatively be affected due not to difference in the DWI signal, but to differences in explicit tissue segmentation as provided to ACT. If the edema is visible in the anatomical image used to derive that segmentation, then whatever tissue component(s) is/are allocated at that location will determine the anatomical priors that ACT applies to streamlines traversing that region. You may alternatively choose to manually allocate that region to the fifth volume in the ACT 5TT image, which will result in no anatomical priors being applied to streamlines as they traverse that region, and therefore whether or not they continue / terminate will be determined exclusively by the input DWI information.

Cheers
Rob

3

Hi Rob,

Thank you for your response. It has been a long time since I asked this question; however, as I am revisiting it, I was wondering if it would be possible to further elaborate on your response.

In particular, I have single-shell DTI data with 30 directions and b-values of 0 and 1000 s/mm². I have performed tractography on this dataset, used the Dehollander algorithm, applied the msmt_csd technique to generate FODs, and utilized mtnormalize. Additionally, I have assigned the entire Edema, Contrast-Enhancing Tissue, and Necrosis regions as the pathology under the 5th volume in the ACT 5tt image.

Could you explain how MRTrix resolves fiber tracts within the Edema region in this scenario?

Additionally, how can tractography be performed specifically within the Edema region alone?

Thanks,
Parand

4

Hi Parand,

This post might help explain the behaviour of streamlines within structures labeled as PATH in the 5TT image.

My understanding is that, unlike the other tissue types, PATH does not impose anatomical constraints on streamlines. More specifically:

  • Unlike CSF, streamlines are allowed to pass through PATH regions rather than being stopped.
  • Unlike SGM, streamlines do not necessarily terminate in PATH tissue - they can continue propagating through it.
  • If FODs are unreliable or missing within PATH regions, streamlines may behave erratically or follow random trajectories.

Hope this helps! :slightly_smiling_face:
Arkiev

5

Dear Parand,

we had some experience trying to perform spherical deconvolution in brain tumors, and are still trying to understand more in-depth whether there exists a proper set of recommendations to follow. Rob gave a comprehensive overview of the issues. Of course, there are alternative methods that don’t use spherical deconvolution (e.g., Abhinav, et al.). More to the point, you might find these two recently published articles to be useful:

The first one uses multi-shell multi-tissue CSD and a set of image processing steps to try to address some of the issues. The second one, based on work from Aerts, et al., combines two CSD methods. Most importantly, we found that the problem was not tracking inside the edema or peritumoral tissue but the 5TT segmentation producing a “gray-matter” boundary preventing any streamline from crossing the lesion. This would mean that, even if we edited the PATH in the 5TT image (as Arkiev correctly suggests), no streamline would be able to enter the lesion in the first place.

I hope some of these resources can be of use.
Best,
Joan

1 Like