Hi Remas,

Not sure about this case, I would suggest to try you identify the "first place" where those NaN values appear as some of the other/subsequent places may be just a consequence of those initial issues. Are those NaN somewhere in the BOLD signal itself? or perhaps they show up first in the first-level covariates created during preprocessing? and if so, in which preprocessing step?, was preprocessing run within CONN or were these covariates imported from fMRIPrep or other software packages?

Best

Alfonso

Originally posted by Remas Alsaadi:

Hello,

I am using CONN Toolbox version 25.b for resting-state fMRI study.

Study design:
- 19 subjects
- 2 conditions:
  - Quran
  - Speaker
- Seed-to-Voxel analysis

I am experiencing persistent NaN-related errors.

Error during second-level analysis:
"Null design matrix (all subjects have missing data or all selected covariates have zero or NaN values)"

Error during denoising:
"Input matrix must not contain NaN or Inf values"

I recreated the project from scratch and reran preprocessing.

I also noticed that some QC covariates contain NaN values. For example:

QC_meanFC = NaN for all subjects 
QC_StdFC = NaN for all subjects
QC_outlierScore = NaN for all subjects 

I tried:
- Removing age covariate
- Excluding one subject that showed 0 value on QC_validsubjects 
- Removing "Effect of Quran" and "Effect of speaker" from denoising

After removing the condition effects, denoising completed successfully, but second-level analyses still produce NaN-related errors.

Could you please advise how to identify the source of these NaN values and whether the issue is related to denoising, QC covariates, ROI data, or project setup?

Thank you very much.