Dear Freesurfer experts,
I am having troubles interpreting the output of the recon-all output concerning the curvature, in particular in relation with the one from the command mris_curvature_stats. My aim is to compare the absolute value of both the gaussian/intrinsic and the mean/extrinsic curvatures between a group of patients and a group of controls. In particular I would like to obtain a voxelwise statistical maps for the comparison.
Browsing the freesurfer mailing list I understood the following (please correct me if I am wrong): - the file $SUBJECTS_DIR/<subjid>/surf/lh.smoothwm.H.crv, produced by mris_curvature_stats, contains the mean/extrinsic curvature map for the left hemisphere, as computed on the white matter surface, for subject <subjid> - the file $SUBJECTS_DIR/<subjid>/surf/lh.curv, produced by recon-all (or by mris_curvature, right?) contains a smoothed version of smoothwm.H.crv - the files $SUBJECTS_DIR/<subjid>/surf/lh.pial.H.crv and $SUBJECTS_DIR/<subjid>/surf/lh.curv.pial are the analogues of the above, but for the pial surface
Also the local gyrification index, contained in the file$SUBJECTS_DIR/<subjid>/surf/lh.pial_lgi, is computed on the pial surface and might be compared (I know it is a different measure) with $SUBJECTS_DIR/<subjid>/surf/lh.pial.H.crv and $SUBJECTS_DIR/<subjid>/surf/lh.curv.pial, not with $SUBJECTS_DIR/<subjid>/surf/lh.curv or $SUBJECTS_DIR/<subjid>/surf/lh.smoothwm.H.crv.
If this is true I expect that: 1- the histograms of the above quantities for any subject be approximately symmetric around zero. Actually given that the convention chosen is to give a positive curvature to the suci and a negative to the gyri (also read in one email of this list from long time ago - so please correct me if I am wrong), I expect that the mean value would lie a bit below zero if we want the brain surface to close (with this convention a sphere has mean curvature -1/r). This expectation of mine is only a rough expectation, but it would be precise if one had normalized somehow for the area related to the vertex on which the local curvature is computed. 2- the histograms for smoothwm.H.crv and pial.H.crv would have fatter tails than curv and curv.pial, respectively 3- smoothwm.H.crv and pial.H.crv would follow quite closely curv and curv.pial, respectively, for each subject
Now please have a look at the attached plots, where in red always appear unsmoothed quantities, while in blue smoothed ones. NB I have extracted the above plotted values using the functions fs_read_Y from the freesurfer longitudinal package by M. Reuter et al.
Now I observe that 1,2,3 are true on the white matter surface (see attached *smoothwm* files). Also, if I average the mean curvature for one subject over all vertices I obtain a negative value. When I look at the analogue plots for the pial surface2,3 are still true, but the mean values for the curvatures are now positive! If I look at the *pial* files, the distribution looks clearly different.
Therefore I have the following doubts/questions: a) is the convention for the curvature sign different on the pial and on the wm surface? b) if not, can the normalization issue I mention in point 1 above explain the negative mean of the mean-curvature I observe (the subject I used to produce the plots is a healthy control, plots are similar for other subjects)? c) in any case I expected the survatures to be more similar to each other on the wm and on the pial surface, given that they are approximately parallel. Any comment on this would be highly appreciated. d) or do you think my results are wrong/ I am not plotting the right thing?
Thanks in advance for any help.
Fabio
Dr. Fabio Bernardoni wiss. Mitarbeiter Psychosoziale Medizin und Entwicklungsneurowissenschaften Tel. +49 (0)351 458-5245 Fax +49 (0)351 458-7206 URL http://www.uniklinikum-dresden.de/psm; www.transdenlab.de Universitätsklinikum Carl Gustav Carus & Medizinische Fakultät an der Technischen Universität Dresden Anstalt des öffentlichen Rechts des Freistaates Sachsen Fetscherstraße 74, 01307 Dresden http://www.uniklinikum-dresden.de Vorstand: Prof. Dr. med. D. M. Albrecht (Sprecher), Wilfried E. B. Winzer Vorsitzender des Aufsichtsrates: Prof. Dr. med. Peter C. Scriba USt.-IDNr.: DE 140 135 217, St.-Nr.: 203 145 03113
Hi Fabio
I'll cc Rudolph who has worked with the curvature much more than I have, but some points:
1. The negative gyral/positive sulcal curvature is a consequence of adopting an outwards pointing surface normal direction.
2. I think the smoothwm.H.crv is generated from the ?h.smoothwm surface, while the ?h.curv is generated from the white. You could generate the equivalent for the white if you wanted.
3. We have 2 different ways of computing curvature. One fits a Hessian to the local height function in the tangent bundle (used in ?h.curv). The other is a discrete method, which I think is used in the smoothwm.H.crv, but Rudolph will know.
4. You can load the ?h.curv files in freeview or tksurfer and verify their appearance.
cheers Bruce
On Tue, 22 Nov 2016, Bernardoni, Fabio wrote:
Dear Freesurfer experts,
I am having troubles interpreting the output of the recon-all output concerning the curvature, in particular in relation with the one from the command mris_curvature_stats. My aim is to compare the absolute value of both the gaussian/intrinsic and the mean/extrinsic curvatures between a group of patients and a group of controls. In particular I would like to obtain a voxelwise statistical maps for the comparison.
Browsing the freesurfer mailing list I understood the following (please correct me if I am wrong):
- the file $SUBJECTS_DIR/<subjid>/surf/lh.smoothwm.H.crv, produced by mris_curvature_stats, contains the mean/extrinsic curvature map for the left hemisphere, as computed on the white matter surface, for subject <subjid>
- the file $SUBJECTS_DIR/<subjid>/surf/lh.curv, produced by recon-all (or by mris_curvature, right?) contains a smoothed version of smoothwm.H.crv
- the files $SUBJECTS_DIR/<subjid>/surf/lh.pial.H.crv and $SUBJECTS_DIR/<subjid>/surf/lh.curv.pial are the analogues of the above, but for the pial surface
Also the local gyrification index, contained in the file$SUBJECTS_DIR/<subjid>/surf/lh.pial_lgi, is computed on the pial surface and might be compared (I know it is a different measure) with $SUBJECTS_DIR/<subjid>/surf/lh.pial.H.crv and $SUBJECTS_DIR/<subjid>/surf/lh.curv.pial, not with $SUBJECTS_DIR/<subjid>/surf/lh.curv or $SUBJECTS_DIR/<subjid>/surf/lh.smoothwm.H.crv.
If this is true I expect that: 1- the histograms of the above quantities for any subject be approximately symmetric around zero. Actually given that the convention chosen is to give a positive curvature to the suci and a negative to the gyri (also read in one email of this list from long time ago - so please correct me if I am wrong), I expect that the mean value would lie a bit below zero if we want the brain surface to close (with this convention a sphere has mean curvature -1/r). This expectation of mine is only a rough expectation, but it would be precise if one had normalized somehow for the area related to the vertex on which the local curvature is computed. 2- the histograms for smoothwm.H.crv and pial.H.crv would have fatter tails than curv and curv.pial, respectively 3- smoothwm.H.crv and pial.H.crv would follow quite closely curv and curv.pial, respectively, for each subject
Now please have a look at the attached plots, where in red always appear unsmoothed quantities, while in blue smoothed ones. NB I have extracted the above plotted values using the functions fs_read_Y from the freesurfer longitudinal package by M. Reuter et al.
Now I observe that 1,2,3 are true on the white matter surface (see attached *smoothwm* files). Also, if I average the mean curvature for one subject over all vertices I obtain a negative value. When I look at the analogue plots for the pial surface2,3 are still true, but the mean values for the curvatures are now positive! If I look at the *pial* files, the distribution looks clearly different.
Therefore I have the following doubts/questions: a) is the convention for the curvature sign different on the pial and on the wm surface? b) if not, can the normalization issue I mention in point 1 above explain the negative mean of the mean-curvature I observe (the subject I used to produce the plots is a healthy control, plots are similar for other subjects)? c) in any case I expected the survatures to be more similar to each other on the wm and on the pial surface, given that they are approximately parallel. Any comment on this would be highly appreciated. d) or do you think my results are wrong/ I am not plotting the right thing?
Thanks in advance for any help.
Fabio
Dr. Fabio Bernardoni wiss. Mitarbeiter Psychosoziale Medizin und Entwicklungsneurowissenschaften Tel. +49 (0)351 458-5245 Fax +49 (0)351 458-7206 URL http://www.uniklinikum-dresden.de/psm; www.transdenlab.de Universitätsklinikum Carl Gustav Carus & Medizinische Fakultät an der Technischen Universität Dresden Anstalt des öffentlichen Rechts des Freistaates Sachsen Fetscherstraße 74, 01307 Dresden http://www.uniklinikum-dresden.de Vorstand: Prof. Dr. med. D. M. Albrecht (Sprecher), Wilfried E. B. Winzer Vorsitzender des Aufsichtsrates: Prof. Dr. med. Peter C. Scriba USt.-IDNr.: DE 140 135 217, St.-Nr.: 203 145 03113
freesurfer@nmr.mgh.harvard.edu
-
Bernardoni, Fabio -
Bruce Fischl