I ran an analysis of a diffusion dataset using pyAFQ. The measures I looked at to quantify potential group differences was the streamline count as well as FA values as part of the tract profiles automatically generated by AFQ. My analysis contains a total of 10 fibre bundles, one of them shows significant group differences in streamline count. In my understanding, this indicates macrostructural differences. Now, I also looked at the tract profiles and it seems that this difference is not reflected in the FA values of the same tract. There are some effects in other tracts though. I am not surprised that there are some microstructural effects in leu of a macrostructural one. However, I was wondering how one should interpret the presence of a macrostructural effect (i.e., difference in streamline counts for a pathway) but absence of a microstructural one in the AFQ tract profile?
Thank for your fast response and the links to the papers!
I understand and can accept the principled shortcoming of streamline counts as discussed in the papers, but I believe they do not really or at least not fully apply to my experiment: I am comparing two groups (think: patients vs. controls) with the identical method, sequence, etc. Plus, I am always comparing the same pathways across the two groups with each other only, not to other pathways.
So, my situation is similar to the “patient with a tumour” example in the Smith et al. paper: The tumour would actually be missed if the number of streamlines was normalized across subjects. Similarly, the fact that I see a big difference in streamline counts in the same pathway between groups should represent an actual biological difference.
Now, I agree that it cannot really be pinpointed what exactly gives rise to the difference due to the general methodological shortcomings of diffusion MRI, etc., but I think it is fair to conclude from this data that there is a group difference that i s not just a methodological artefact. And this brings me full circle to my initial question, what it would mean to see an effect when looking at streamline counts, but not observing an effect in the same pathway when looking at its microstructure (i.e., FA)?
Streamline count has a ton of biases related to streamline length and ROI size. It also doesn’t necessarily relate to tract volume (as you can have lots of streamlines dense in a small area, or the same number of streamlines that are not dense taking up a larger area).
Also, regardless of what macrostructure is used, it is not necessarily the case that microstructure and macrostructure are related. But, seeing how different microstructural metrics change in relation to one another could be helpful, see here: How to interpret dMRI metrics | DSI Studio Documentation
Thanks, @Steven for you answer and the resources. As said, I acknowledge that streamline count has a ton of biases, but none of these biases would lead to a group difference along the lines I described. Or at least I don’t see any way in which a methodological problem could cause the difference here.
Accordingly, my question as to what the different results mean is not really answered by that, but I guess the best guess is that macro- and microstructure are just not necessarily related.