help > Zeros in PCAcov (MVPA)
Mar 14, 2025  05:03 AM | paula faimann - unimelb
Zeros in PCAcov (MVPA)

Dear CONN experts,


I am running an MVPA analysis and want to calculate the total variance explained by my components. I understand that the PCAcov file contains the percentage of variance explained by each voxel, and I would like to determine the total variance explained by the components.


The PCAcov files contain a 3D matrix (91×109×91). 242545 voxels have nonzero values, while 660084 are exactly zero. I can compute the mean across all voxels for each component (I know the variance explained is cumulative), but I am unsure whether to:



  1. Include all values (including zeros) when computing the mean.
  2. Exclude zero values, assuming they represent non-brain or masked voxels.

When I include zeros in the first component, I get a value of 5%, but when I exclude them, the value increases to 17%.


How should I interpret the zero values in this context? Does PCA in CONN assign zeros to voxels that do not belong to any component? 

Moreover, is there any chance I could see a mask of the voxels that contribute to each component?

I appreciate any insights you can provide.


Best regards,
Paula