I was running pydeface on a dataset, and had a few subjects with imperfect defacing. I tracked down the problem to the subjects' lower image quality causing bad registration to the standard mean_reg2mean.nii.gz, no matter which cost function I use. Therefore, I manually ran registration of the structural to standard MNI 1mm brain provided in $FSLDIR/data/standard, and applied the inverse transform to get standard -> subject matrix. However, the facemask still was not aligned properly. Long story short, I pulled up the mean_reg2mean.nii.gz provided by pydeface and overlaid the standard MNI152_T1_1mm.nii.gz, and found out they do not appear aligned! I tried a few viewers including fsleyes, mricron, and overlaid a bunch of different MNI standard images, and got the same result. I finally re-ran the registration using skull-stripped versions of the structural and mean_reg2mean.nii.gz, and got a good mask. Thought I would check in to make sure I am not missing anything obvious. Image attached.
I understand this may seem like a minor issue, since pydeface runs registration via flirt anyways. However, having the facemask.nii.gz in actual standard space would allow the --template option to use any MNI standard space image, and automatically have facemask.nii.gz work, without the user needing to supply a custom facemask. A small fix for a lot of benefit, I would say.
P.S., would also be nice to be able to load transform matrices and skip registration altogether if the user has a working transform matrix, saving a lot of time.
I was running pydeface on a dataset, and had a few subjects with imperfect defacing. I tracked down the problem to the subjects' lower image quality causing bad registration to the standard mean_reg2mean.nii.gz, no matter which cost function I use. Therefore, I manually ran registration of the structural to standard MNI 1mm brain provided in $FSLDIR/data/standard, and applied the inverse transform to get standard -> subject matrix. However, the facemask still was not aligned properly. Long story short, I pulled up the mean_reg2mean.nii.gz provided by pydeface and overlaid the standard MNI152_T1_1mm.nii.gz, and found out they do not appear aligned! I tried a few viewers including fsleyes, mricron, and overlaid a bunch of different MNI standard images, and got the same result. I finally re-ran the registration using skull-stripped versions of the structural and mean_reg2mean.nii.gz, and got a good mask. Thought I would check in to make sure I am not missing anything obvious. Image attached.
I understand this may seem like a minor issue, since pydeface runs registration via flirt anyways. However, having the facemask.nii.gz in actual standard space would allow the --template option to use any MNI standard space image, and automatically have facemask.nii.gz work, without the user needing to supply a custom facemask. A small fix for a lot of benefit, I would say.
P.S., would also be nice to be able to load transform matrices and skip registration altogether if the user has a working transform matrix, saving a lot of time.