Hello dear MRtrix community,
I am wondering whether mrview has a function to display only fixels which clusters above a certain threshold.
To be more precise, in SPM-fMRI you can filter out single voxels of activity because they are too small, and have a “clean” output figure. Can mrview do the same for fixel plots?
Thank you very much,
Giuseppe
No such functionality within mrview itself, no. But if you’re looking at the results of fixel-based analysis, then you’re looking at threshold-free cluster-enhanced results, so they’ve effectively already had that kind of filtering applied.
A filter like you describe might be of use if you’re looking at different types of fixel-based outputs that are statistical in nature, yet not produced using our connectivity-based fixel enhancement (CFE) – in which case it would be interesting to hear of such use cases!
Hi Giuseppe,
The concept of a “cluster” is not as simple with fixel data as it is for voxel data. There are however a couple of options:
Map fixels to voxels, perform a connected-component analysis, omit small components, and then take the intersection of that mask with the original fixel mask.
fixelfilter does not exclusively perform fixel-based data smoothing: there is also connect, which is a basic attempt at a fixel-based connected-component analysis utilising the fixel-fixel connectivity matrix. It is moreso intended for the separation of discrete isolated bundles rather than the rejection of spatially disconnected fixels (as two fixels that are two voxels apart may still possess a high fraction of connectivity to one another), but it’s nevertheless there if you want to experiment with it, and might have utility if combined with the first point above.
The TSF-based smoothing described in the Wiki page can potentially reduce the prevalence of disconnected statistically significant elements when visualised using streamlines rather than the fixels themselves.
Rob
Hi,
Sorry to revive an old thread, but I am beta testing an R module (GitHub - PennLINC/ModelArray: R package for running statistical models on Fixel data) that can make fixel-metric inferences and does not use CFE (at least for p-val correction, but CFE-smoothed images are used). While this package has some unique features (lower memory requirements, generalized additive modelling capabilities), the current p-val correction is independent FDR, which may be strict as it does not account for dependence between neighboring fixels in a fiber population. It also results in a number of small hard-to-interpret fixel clusters like the original poster described.
I would be curious if the p-val permutation FWE approach used in CFE could be adapted for this R package, or if at the very least there is a standalone function that one could use to perform this permutation post-hoc, given uncorrected p-vals.
Thanks,
Steven