Hi Jen

Yes, you are perfectly right, connectivity estimates in subjects with very low degrees of freedom will tend to be very noisy/unreliable so they are often used as part of your exclusion criteria. In CONN I recommend using a dataset-specific threshold rather than an absolute threshold (i.e. subjects with dof below the first quartile minus 3 times the interquartile range). If you are using the latest version of CONN this will be in fact part of the criteria used by CONN to propose your dataset inclusion/exclusion set in the Denoising&QC tab (see example image attached for where to find that information). That said, using an absolute threshold like 10 or 20 is also a perfectly reasonable strategy.

Hope this helps

Alfonso

Originally posted by Jen K:

Hello,

I recently completed the preprocessing and denoising steps for my pediatric dataset and noticed that some subjects have very low degrees of freedom after denoising. The original degrees of freedom were 163, but for some subjects they drop to fewer than 10.

My understanding is that this reflects a limited amount of usable data. I am therefore considering excluding subjects with very low degrees of freedom. But I was wondering whether there is a recommended cutoff for degrees of freedom when excluding subjects, or whether exclusion should instead be based on a different metric.

Thank you!