Dear Experts,
This is more a conceptual question. I have reconstructed some basal ganglia pathways in several Parkinson’s disease patients. One possible bias is that Parkinson is a degenerative disease. Therefore it is expected to have an altered diffusion signal in these patients. This could, theoretically, affect the reconstruction of the basal ganglia pathways. Using the SIFT algorithm would account for this possible bias?
Thank you!
The answer to your question is probably yes, but depending on exactly what you want to do with the data, and how you measure any differences, different steps may be required. First off, you’ll need to use some form of bias field correction in any case. Next, if you intend using inherently normalised graph measures, then SIFT alone should be fine. If you intend somehow normalising your connectome matrices (e.g. to have the same overall density), then again, SIFT alone should be fine. If however you intend to perform analyses where the actual post-SIFT values are compared directly, e.g. by comparing post-SIFT streamlines count (or sum of streamline weights if using SIFT2 – recommended), then all the considerations relevant to an apparent fibre density analysis also apply (it’s the same underlying model), particularly those that relate to global intensity normalisation. So maybe your best bet is to tell us a bit more about what you plan on doing, and then we might be able to point you to the most relevant recommendations.
Hi @jdtournier, Thank you so much for your answer. I have 43 Parkinson’s disease (PD) patients and I created a FOD population template using these patients. I want to reconstruct some basal ganglia pathways using this template (I am using the template because the study design includes the analysis of PD patients as a group).
I am assuming that the reconstructed pathways would be, in some way, representative of all those PD patients. Then I would like to 1.- quantify the streamlines count (or sum of streamline weights using SIFT2), 2.- compare the streamline count between different basal ganglia pathways and 3.- correlate the streamlines count with some clinical information. As you pointed out, using SIFT, I am assuming that the streamline count would have a biological significance taking into account the white matter degeneration occurring in PD patients.
Thank you so much for your help!!
My apologies, I’m flat out with the last minute preparations for the ISMRM and next week’s workshop…
To briefly answer your question, there’s several conceivable approaches to do this:
You could perform this in native space per subject, by defining some robust ROIs to delineate the pathway in some template space, and warping these ROIs onto each subject independently. Then extract the pathway of interest with sufficient density, but also perform a whole-brain tractogram, and merge the two using tckedit. At this point, you can run tcksift2 on the combined tractogram to obtain the streamline weights, and then use tck2connectome to extract the sum of streamline weights for the pathway of interest (by defining relevant ROIs as ‘nodes’). At this point, the sum of streamlines weights should represent the information you’re after. That way, you can also use the -act option in tckgen to help remove false positives.
The other way is more in line with what you suggest. There is a command in MRtrix3 called afdconnectivity, which superficially does exactly what you want. But there is a very real possibility that the voxels from which the AFD is estimated also contain contributions from other pathways, which would contaminate the estimate. There is a new -wbft option to that command that might help here, if you can produce a whole-brain tractogram and extract your pathway of interest from that – I’m not familiar with how it works exactly, but at a guess it would require a dense whole-brain tractogram, filtered using tcksift, and then tckedit to extract the subset of interest representing your pathway. This could all be performed in template space, which makes things easier but also adds the potential confound of interpolation errors during the registration, and would require modulation of the FODs during warping (@maxpietsch, is this still possible…?).
I could probably help a bit more in a couple of weeks, but that’s all I have time for right this minute – sorry!
Not using mrregister (never was possible afaik) but ODF modulation can be applied using mrtransform (docs).