This MR fixes a few bugs that happened when the voxel-wise design option is used. In brief,
- Load the voxel-wise design after loading the data and doing the masking. Before, it was done prior this and could be problematic because the mask was not yet processed.
- The WB p-value are now computed within the loop over contrast. Before, they were computed outside, which was not correct.
- Fix a bug where some F-scores would not be computed using the appropriate SwE.
- Fix a bug where the cluster extent and the cluster mass images would not be saved correctly
- Add a missing multiplicative factor in the computation of the number of degrees of freedom in one of the method
- Now, save the number of degrees of freedom if requested
- Reactivate the voxel-wise design option (though it is still hidden from the help).
This MR was tested by compararing the results of the voxel-wise design option with a covariate replaced by minus itself in approximatively half of the voxels. The results seem consistent with those obtained without using this voxel-wise design option. I could not find any issue left.
Before merging, it seems that the we should show the option again in the help.