I am processing multi-echo block-design task fMRI data with tedana. I have extracted activation statistics using both the high kappa (hik_ts_OC.nii) dataset and the denoised (dn_ts_OC.nii) dataset and get higher z-scores using the hik_ts_OC.nii. I have two questions:

Is it appropriate to use the hik_ts_OC.nii dataset to extract activation statistics?

If so, how does that affect the degrees of freedom?

The high-kappa time series only includes accepted components (i.e., those with BOLD-like profiles), while the denoised time series includes both accepted as well as ignored components (i.e., those with profiles reflecting thermal or other Gaussian noise, or explaining a low percentage of the total variance).

We generally do not recommend using the high-kappa time series in subsequent analyses, exactly because it has a substantially lower number of degrees of freedom than the denoised time series. You should be able to see exactly how many degrees of freedom are in each time series by looking at the ica_decomposition.json file, which provides labels for each ICA component.

Hopefully this is helpful ! Let me know if you have any other questions, or if you have any suggestions for how we could improve our description of the tedana outputs.

That is pretty much what I thought. Just so I’m 100% clear, if a dataset has 20 accepted components, the high-kappa time series has 20 degrees of freedom? On the other hand, the denoised time series will have (number of timepoints - number of rejected components) degrees of freedom?