Hi Paul
yes, the -mprage flag could definitely cause what you are seeing. It assume higher CNR and lower SNR in the images (hence bigger class separation between gray and white), and if used on SPGR it could definitely erode boundaries like this.
cheers Bruce On Fri, 31 Jan 2020, Wright, Paul wrote:
External Email - Use Caution
Dear FreeSurfer experts
I am a relative novice with FreeSurfer. I am reprocessing some older images with v6.0 that were previously processed with v5.3. In the newer batch, the GM-WM boundary is wider and the pial boundary is pushed out, but the newer version does a better job in the hippocampus. I am trying to figure out what parameters might differ between versions so I can get a cortical reconstruction as well as hippocampal segmentation. Both batches used plain vanilla recon-all -all.
The attached figure shows v5.3 on top and v6.0 below, with aseg.mgz over T1.mgz. I used an example that as well as thin cortex in the new version, shows a hippocampal error in the older version. (The slices are aligned at the hippocampus, but the pitch is slightly different so they don't line up more superiorly.)
From the T1, it looks like the step that sets white matter to 110 is more
aggressive in 6.0. I looked through recon-all.cmd and noticed both normalization steps include the -mprage flag in 6.0 not 5.3.Could this be the why the results differ? (The images are SPGR not MPRAGE.) Looking at help for mri_normalize it doesn't mention this flag or how to switch it off, so I'm stuck here. Of course, it might be something else, not the -mprage flag.
I'd much appreciate any guidance from the list.
Best wishes Paul
Paul Wright, PhD Postdoctoral Researcher, Department of Neuroimaging
King's College London Maurice Wohl Clinical Neuroscience Institute 125 Coldharbour Lane London SE5 8NU
Desk: +44 20 7848 8579 Mobile: +44 7491 905754 Email: p.wright@kcl.ac.uk