neuroquant / fmri-site-effects Goto Github PK

Automatically score the presence of site effect on the time-series data

Include a range of subjects (10-15 per site, per scanner (if there is scanner))

• Separately code scanner effect
• Separately code site effect

This will need to involve Schott test; multi-level covariance models.

Given the nice set of QA measures available in the ABIDE data

Idea: I think it would be interesting to compare site-to-site differences in different QA measures --- as an index of site effects stemming from physiological/nuisance artifiacts.

http://preprocessed-connectomes-project.org/abide/quality_assessment.html

If we have these, we should be able to compare if better site effect correction reduces how much subject to subject global connectivity is correlated with these artifiacts. [As a very elementary thing to check/illustrate with a few lines of code]

From tutorial/pedagogial perspective: This both serves as an important motivation/need for site effect correction in multi-site studies as well as index of reducing such confounds. Very like how powers does pure motion to connectivity correlation.

Incorporate I2C2 within the multivariate CCC calculations with a 'ccc' mode or 'icc' mode.