Dear Experts,
I have some data which was acquired in low-resolution: 4 mm isotropic. Standard preprocessing would lead to 2mm isotropic images but that seems like a massive amount of upsampling. My goal is to use these images for MVPA - I understand that using a 4mm normalization, rather than 2mm, would be helpful because of lower space requirement and fewer features (so speed). One of the reviewers raises an argument that normalization to 2mm would lead to more correlated voxels which might make an impact on the performance of ML algorithms.
Are there any reasons (other than space/time/convenience) to not upsample to regular 2mm but stick to acquisition voxel size during normalization? Are there papers that have studied the impact of spatial normalization voxel size on fMRI?
Thanks
PB
|