no parcellation for GLM with Freesurfer. #343

Activity

1. knutj commented on Mar 17, 2023

   

   knutj

   on Mar 17, 2023

   Author

   More actions

   Can you also please add Desikan-Killiany Atlas from Freesurfer to fetch parcellation.
2. zihuaihuai commented on May 4, 2026

   

   zihuaihuai

   on May 4, 2026

   Collaborator

   More actions

   Hi @knutj — a couple of things bundled here, let me address them separately.

   1. GLM on FreeSurfer .mgh data with parcellation output (MATLAB). You can use BrainStat's MATLAB stack directly; the input just needs to be an n_subjects × n_vertices matrix on fsaverage (or whichever template you used). Rough sketch:

   % Load thickness for each subject onto fsaverage
   Y = zeros(numel(subjects), 327684); % both hemispheres, fsaverage
   for i = 1:numel(subjects)
       [lh, ~] = SurfStatReadData1(fullfile(subjects{i}, 'surf', 'lh.thickness.fsaverage.mgh'));
       [rh, ~] = SurfStatReadData1(fullfile(subjects{i}, 'surf', 'rh.thickness.fsaverage.mgh'));
       Y(i, :) = [lh; rh]';
   end
   
   % Build the model
   age   = FixedEffect(tbl.age);
   sex   = FixedEffect(tbl.sex);
   group = FixedEffect(tbl.group);
   M     = 1 + age + sex + group;
   contrast = tbl.group == 'patient';
   
   % Fit, with FDR correction
   slm = SLM(M, contrast, 'surf', 'fsaverage', 'correction', {'fdr'});
   slm.fit(Y);
   
   % Per-vertex p-values are in slm.Q. To roll them up to a parcellation:
   parc = fetch_parcellation('fsaverage', 'schaefer', 400);  % or 'aparc' for DK
   roi_mean_p = accumarray(parc(parc>0), slm.Q(parc>0), [], @mean);

   The "parcellation for each p-value" part isn't a built-in helper — you aggregate the per-vertex slm.Q (FDR-corrected) over whichever parcellation you load.

   2. Adding the Desikan–Killiany (aparc) atlas to fetch_parcellation. Reasonable ask; today fetch_parcellation ships Schaefer / Glasser / Yeo / economo. DK is already on fsaverage as ?h.aparc.annot so it would be a small loader addition rather than a new download. PRs welcome — the relevant function is brainstat.datasets.parcellation.fetch_parcellation (Python) and fetch_parcellation.m in the MATLAB tree.

   3. MATLAB tutorial gap. Confirmed — the Sphinx tutorials are Python-only; MATLAB users currently have to translate by reading the function help. A short MATLAB walkthrough alongside the Python ones would be a useful addition; happy to take PRs.
3. zihuaihuai

   added

   DocumentationImprovements or additions to documentationImprovements or additions to documentation

   MATLABOn the MATLAB implementationOn the MATLAB implementation

   questionFurther information is requestedFurther information is requested

   on May 4, 2026
4. zihuaihuai commented on May 4, 2026

   

   zihuaihuai

   on May 4, 2026

   Collaborator

   More actions

   Partially addressed in #380: fetch_parcellation(template, "destrieux", ...) now returns the FreeSurfer aparc.a2009s parcellation via nilearn (fsaverage5 native; other fsaverage templates resampled nearest-neighbour). The original ask also included aparc/Desikan-Killiany; that one needs a hosted .annot file, so it's left as a follow-up tracked here. The MATLAB-tutorial gap from your third point is still open and would be a useful contribution.
5. zihuaihuai

   added a commit that references this issue on May 4, 2026

   Merge pull request #380 from MICA-MNI/fix/issues-274-341-343

   ...

   8a76a72
6. zihuaihuai

   closed this as completedin #380on May 4, 2026