correlational tractography number of tracts guideline

[Mark Vernon]

Hello,

After running correlational tractography on a database, I get ~88,000 tracts as a result.  After filtering by an ROI, this number is reduced to anywhere from 200 - 4,000 tracts depending on the ROI.  Then, I Recognize and Bundle the filtered tracts which groups the filtered tracts into bundles ranging from several hundred tracts down to a single tract for a bundle.  I'd like to run some stats on the average QA of some of these bundles. My questions are:

1. Are there any guidelines for how big a bundle (number of tracts in the bundle) should be for stats to be meaningful? i.e. a bundle made up of a single tract may not be as helpful but perhaps one with 17 tracts would?

2. Would these 1 or 2 tract bundles simply need to be trimmed off?  If so, are there any guides on how to trim bundles? Particulary, what is part of a bundle and what is not, for someone who is not as experienced in the neuroanatomy of these WM bundles?

Thanks so much!

Mark

[Fang-Cheng Yeh]

I am sorry for the late reply. Here's my responses.

1. Generally, there is no set guideline for how big a bundle needs to
be for statistics to be meaningful. The sample size needed for
statistical significance depends on a variety of factors, including
the effect size, variability in the data, and the desired level of
significance. In general, larger sample sizes (e.g. > 100) are
preferred, but statistical tests can still provide meaningful results
even with small sample sizes. However, smaller sample sizes can be
more susceptible to outliers and other sources of variability, so it's
important to use caution when interpreting results from smaller
bundles.

2. Whether to trim small bundles or not depends on the research
question and the goals of the analysis. If the focus is on larger,
well-established white matter bundles, then it might make sense to
exclude small, potentially noisy bundles from the analysis. On the
other hand, if the goal is to identify smaller, less well-defined
pathways, then it may be important to include smaller bundles in the
analysis.

Best regards,
Frank