Hello FSL users and experts,

We’re using Probtrackx to resolve white matter tracts in a large participant sample. In most cases, our goal is to resolve homologous tracts (e.g. left and right SLF) connecting between seed masks generated from a previously published fMRI meta-analysis. Our initial results using Probtrackx suggest that there is a great deal of variability in the fdt_paths image, even within-subjects, between hemispheres, and even after accounting for total streamlines sent out (e.g. fdt_paths/total streamlines). This is an issue with our younger adult subjects, and expect that it will only be exacerbated when we evaluate older adult subjects. For example, in one participant, and at a common threshold, the left SLF will be quite robust and the right SLF quite weak – in the next subject the opposite is true, and in the next, at that same threshold there are no left SLF results surviving (only visible at a lower threshold).  We understand that there are several factors at play here. But, we wanted to float an idea by the developers and community for handling individual and tract-specific differences in such ‘trackability’ issues.

Our overall goal is to be able to threshold tractography results at a common threshold across tractography analyses (e.g. the genu, SLF, ILF, etc), hemispheres (e.g. left and right SLF), and subjects.  This might be a tall order, but we would like feedback on the approach we outline below.

We’re running these analyses in –network mode, so the tractography is proceeding in both directions between two seed regions in each analysis. First, after probtrackx we’d proportionalize the fdt_paths image by the total streamlines (e.g. streams attempted x total seed voxels in native space). This is an image where we are currently seeing quite variable results in terms of robustness across analyses (a common threshold yields a wide range of results across tracts, hemispheres, subjects). It’s useful that it is in proportion scaling rather than successful streamlines – but, again, is quite variable across analyses. This is likely due to factors that affect the robustness of the analysis (e.g. ROI group-to-native registration alignment, image quality, caliber/volumetric differences, and so on). We suspect that scaling this proportion image by an estimate of ‘success’ of the tractography would help to place each participant’s results into a more consistent range.  We’d like to integrate waytotal as a marker of ‘success’ here. In order to handicap less ‘successful’ analyses we’d, next, re-scale waytotal across all tractography analyses performed (to a range of 0-1, with 1 representing the highest waytotal in the study).  After dividing the proportionalized fdt_paths image (fdt_paths/total_streamlines) by the re-scaled waytotal (waytotal_max[across all analyses]/waytotal[current analysis]) we believe we’d have a reasonably adjusted proportionalized fdt_paths image to which a consistent threshold could be applied.

Does anyone feel like we’re missing/overlooking anything here? Waytotal is influenced by more than just trackability, more streamlines sent out = higher potential waytotal so we see there might be an issue using it as a scalar to an image proportionalized by total streamlines. Are we placing too high a value on a common threshold across analyses?  Our over-arching goal is to threshold tractography results in native space and summarize diffusion parameters within these thresholded tracts. But, we fear thresholding one tract/one hemisphere/participant by one value and another by a second value (and so on) based on an arbitrary heuristic (‘it looks good’) is grounds for criticism (e.g. investigator degrees of freedom).

We appreciate any feedback!