Interpreting NaN values in IDP files for developing QC IDP thresholds

[dsodums]

I was examining the IDP values for the Rockland dataset outputs and was hoping to get more information about what the following variables represent in terms of NaNs and if this is of concern or not if we do have lines in the SC or TL that have NaNs.

SC_nan_lines
SC_num_nan

TL_nan_lines
TL_num_nan

For example, line 205 is often an NaN in many of our outputs, I'm assuming this means that there might have been some problem with the structural processing such that ROI #205 is problematic.

We are looking to develop thresholds to apply semi-automated QC so we have processed a subset of ~200 cases and manually QC'ed them, similar to what was done with the CAM-Can dataset in the tvb-ukbb paper, and I'm hoping to confirm this potential issue before we start running any analyses with the QC 1-5 indices and the IDP values.

[justinliuyuwang]

Hi Devin!

num_lines are the ROI #s of all-NaN lines
num_nan is the number of all-NaN lines

There may have been an error with tractography from/to ROI 205 if you are getting a NaN row. Depending on your end goals for the dataset, you may need to resolve this issue. If there are no obvious errors in structural and diffusion processing, then trying a different parcellation may be helpful.

[justinliuyuwang]

You are missing a closing bracket after num_nanlines!

[dsodums]

Ah that's it, thanks so much!

In your experience, how concerned would your group be about a few NaNs in SC/TL if the FC matrices themselves aren't showing any NaNs?

[justinliuyuwang]

This is a difficult question to answer as it depends on your project needs. A few NaNs here and there may not be a huge point of concern, but how you treat nans in analysis is important. If you accept matrices with nans those nans should not be included in analysis.

[dsodums]

That makes total sense, many thanks!

Would it be of interest to this group for us to post what we plan to do with the outputs in a separate discussion thread?

[noahfl]

Yeah absolutely! Any non-bug-related discussion about the pipeline, how people use it, approaches, etc. is more than welcome 🙂