Hi Chris et al.,

I've recently used PALM with various 4D .mgh input files (i.e., subjects' mri_concat[enated] lh.thickness.fwhm15.fsaverage.mgh files, for example), so PALM's results are in .mgz format. I can convert these to .nii using mri_convert in order to use fslmaths/fslstats and load them into Surf Ice, but the overlay is a uniform map with (what appears to be) 0 at all vertices. In case it's relevant and you're familiar with PALM, I've generally used the -logp option in order to save resulting p-values on a -log10p scale, as suggested by Anderson Winkler, but perhaps this complicates things for Surf Ice. I've run into various quirks loading results in Freeview, which has partially led me to explore Surf Ice, but the maps are generally interpretable and highly significant as viewed in Freeview. For example, there should be three significant clusters in the attached screenshot of one .nii overlay (inferior precentral, superior parietal, and inferior parietal).


Do you have any suggestions on how I might go about either 1) properly loading these results in Surf Ice or 2) modify the PALM input or output files so that Surf Ice can correctly display these results?

Many thanks for your help, and thank you for creating such a beautiful, user-friendly rendering tool for the neuroscience community! It's truly beautiful :-)


With gratitude,
Dan

Dan-

This might be a better question for Anderson. The crucial thing about any statistical overlay is ensuring that there is a good spatial correspondence between the background image and the overlay. If these do not match, the results will not be impressive. It is worth noting that FreeSurfer typically uses its own template images, and that the different levels of inflation break the spatial correspondence: e.g. when you blow up a surface to a 2D flatmap it no longer retains the same spatial position as your input image. Surfice assumes you have taken care of this yourself. I wonder if your current images are better aligned to the pial map, not the inflated map you show. If you are using a recent version of Surfice, it should read mgh/mgz files directly, so that would avoid the mri_convert. 

1. I would pay special attention to the relevant wiki section  - perhaps Darren's mni2fs or freesurfer itself can help maps the 3D voxels to your desired 3D image, and you can overlay the same image.

2. Given that your data was analyzed as a 3D volume rather than a surface, you may want to show the activation as a 3D mesh. As long as your background image and overlay have the same alignment (which is required of any method for overlay), this should eliminate the challenge of warping a 3D value onto a 2D surface. You will need to choose a nice statistical threshold. I tend to like z-values rather than logP, but those are easy to convert between.

3. Given that your input data is already a volume, I would consider if a mesh is the best way to show this. I tend to like mesh colors if the analyses were done on surface measures, but you need to take great care when transforming between these representations. This might be a good application fro MRIcroGL instead of Surfice.

OK, so PALM uses the MGZ/MGH format for per-vertex scalar intensity values, rather than as 3D images. The upcoming release of Surfice will handle these. I also think an upcoming version of PALM should include new features that will ease these issues.