Using customized stats code for mass-univariate approach in freesurfer for structural data
Hi - If I would like to try to perform mass-univariate statistics at the vertex level (thickness, area, vol at each vertex) without using the mri_glmfit, is this order a sound step (in general) for mapping stats parameter of choice (p-val or t s or weights for example) to the surface?
1. Use mris_preproc and surf2surf to get the data to common space and smooth.
2. Load the data to matlab using fs_read_Y() or MRIread
3. Perform stats at each vertex using stats toolbox to apply modeling approach of your choice (e.g., machine learning, linear mixed model etc.)
4. Extract P-values (or other parameters) for each vertex for the variable of primary interest
5. Use MRIwrite to create mgh file
6. Perform multiple comparison using freesurfer functions (e.g., FDR or monte carlo sim)
7. view corrected maps using tksufer?
Any steps I should consider adding or avoid?
Thank you!
Hi Yoonho,
the steps look good. Take a look at https://surfer.nmr.mgh.harvard.edu/fswiki/LinearMixedEffectsModels which basically describes exactly these steps for the case of linear mixed effects models. Note that we have dedicated code for that as part of FreeSurfer (see also the references explaining why this code is great, e.g. "Spatiotemporal Linear .... Neuroimage 2013")
Also, you can do a more powerful 2stage FDR correction using lme_mass_FDR2 in matlab directly (best to combine hemisphers for that, see wiki).
cheers, Martin
On 05/24/2016 01:55 PM, Chung, Yoonho wrote:
Hi - If I would like to try to perform mass-univariate statistics at the vertex level (thickness, area, vol at each vertex) without using the mri_glmfit, is this order a sound step (in general) for mapping stats parameter of choice (p-val or t s or weights for example) to the surface?
- Use mris_preproc and surf2surf to get the data to common space and
smooth.
Load the data to matlab using fs_read_Y() or MRIread
Perform stats at each vertex using stats toolbox to apply modeling
approach of your choice (e.g., machine learning, linear mixed model etc.)
- Extract P-values (or other parameters) for each vertex for the
variable of primary interest
Use MRIwrite to create mgh file
Perform multiple comparison using freesurfer functions (e.g.,
FDR or monte carlo sim)
- view corrected maps using tksufer?
Any steps I should consider adding or avoid?
Thank you!
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
That looks fine. If you are going to use the monte carlo sim, then you'll need an estimate of the FWHM. In mri_glmfit, we get that by running mris_fwhm on the residuals. So, assuming that your analysis has residuals, then you should save those out too.
On 05/24/2016 01:55 PM, Chung, Yoonho wrote:
Hi - If I would like to try to perform mass-univariate statistics at the vertex level (thickness, area, vol at each vertex) without using the mri_glmfit, is this order a sound step (in general) for mapping stats parameter of choice (p-val or t s or weights for example) to the surface?
- Use mris_preproc and surf2surf to get the data to common space and
smooth.
Load the data to matlab using fs_read_Y() or MRIread
Perform stats at each vertex using stats toolbox to apply modeling
approach of your choice (e.g., machine learning, linear mixed model etc.)
- Extract P-values (or other parameters) for each vertex for the
variable of primary interest
Use MRIwrite to create mgh file
Perform multiple comparison using freesurfer functions (e.g.,
FDR or monte carlo sim)
- view corrected maps using tksufer?
Any steps I should consider adding or avoid?
Thank you!
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu
-
Chung, Yoonho -
Douglas N Greve -
Martin Reuter