Manually extracted tract comparison

Good afternoon, I have two groups of subjects (10 control x 10 patients) and I would like to compare manually extracted U-fibers that connect the ventral pre and postcentral gyrus. I have some questions regarding this method:

  1. I have seen it in other papers, but I wonder whether this methodology is valid compared to TBSS or fixel-based analysis.

  2. I would like to know if any tract metric could be used in this comparison, including FA, MD, tract volume and number of streamlines.

  3. Finally, would it be necessary to draw ROIs with the same numbers of voxels and select the same number of streamlines for each subject in order to make this comparison possible?

Thank you


Hi Thiago,

  1. I think that a greater amount of detail on the “methodology” would be necessary to determine whether or not it is valid; there’s a lot of potential wiggle room here, especially if there is manual delineation involved.
    Also, I don’t know if one methodology can be valid “compared” to another; it is either valid or not in its own right. A comparison to other methods would be more about sensitivity to effects & interpretability of results. While streamlines-based quantification for endpoint-to-endpoint connection strength makes sense from a theoretical perspective, there is always the concern of increased variance in the data due to the unreliability of streamlines tractography, which TBSS and FBA are independent / less dependent on.
    Given you mention voxel-wise metrics, it’s also worth being aware of another alternative where the tract is delineated in some template space, and the resulting mask is then transformed back to the space of each individual participant, for determining the voxels from which to sample the metric.

  2. There’s nothing stopping you from using any of those metrics; obviously each has their pros and cons. Tract volume is sensitive to length, which may not be desirable. As far as number of streamlines is concerned, there are lots of review articles that explain why this is not a great metric; I’ll also refer to you my preprint that explains the proposed solution.

  3. The answer here depends on the decisions regarding the prior questions. E.g. Obviously one would not select the same number of streamlines for each subject and then perform statistical inference on the number of streamlines. Given the breadth of possibilities here, I can only suggest thinking very critically about what the consequences of each of those decisions would be in the presence of individual variability, and whether there are more objective mechanisms available for determining which voxels / streamlines are / are not attributed to that pathway in each subject.