[ENH] cluster threshold for surface GLM

[mszinte]

Is there an existing issue for this?

• [X] I have searched the existing issues

Describe your proposed enhancement in detail.

As I use these days mostly (if not only) surface analysis of fMRI data, I use to the surface GLM analysis of nilearn, however I couldn't find the cluster_threshold method available for volume analysis. The idea of cluster analysis for surface would be to discard isolated vertices from the GLM output in order to generate thresholded maps with small cluster of vertices removed following a cluster_threshold argument. Here is the volume based equivalent: https://github.com/nilearn/nilearn/blob/436cfb2196b69054d876e592a59c5a391b36f4c8/nilearn/glm/thresholding.py#L184 After early discussion I feel that for this improvement (@Remi-Gau @bthirion, @ulascombes) , we would need first:

• to associate each vertex to an area
• define connection component of each vertex
• set rules of connection of a vertices (maybe geodesic distances)

Benefits to the change

This analysis is to my opinion an important missing step for the basic GLM surface analysis of nilearn, it will benefit to the community of users who moved toward fMRI surface analyses.

Pseudocode for the new behavior, if applicable

# insert your code below

[bthirion]

Well, you don't want to have to compute geodesic distances to obtain connected clusters. You simply need to use the topology of the mesh (corresponding to the face definition). But I agree that this would be useful.