Have there been recent changes to how CONN handles structural segmentation and normalization?  I've noticed, using the same dataset, that segmentation and normalization came out better in earlier CONN versions (16 and 17), than in the most recent iterations.  The problems with the newer gray and white segments have been so bad, that I've gotten in the habit of using CAT12 segmentation to generate my tissue maps and then replace those CAT12 derived maps with the ones that CONN accomplished.

Here are two examples of poor segmentation output using CONN 20.

Thx and warm regards, 
Jeff

Problem with Gray/CSF.

Hi Jeff,

Thanks for bring this up. While there have been minor differences in SPM12 normalization procedure across different SPM12 releases, and also minor changes in how CONN uses normalization (e.g. functional normalization uses ART outlier-free mean functional since version 17c, output files are resampled at 1mm instead of 2mm voxels since 18a) none of these would explain such a drastic difference in the result (and I have also not experienced any drop in the quality of the outputs). Could you perhaps send me the original structural file in these examples (avg_INT098_T1_baseline.nii) to see if that helps me replicate and better understand this issue?

Thanks again!
Alfonso
Originally posted by Jeff Browndyke: