fmri_cleanclusters

Description

FMRI_CLEANCLUSTERS Removes small and sparse clusters from a thresholded map. Applies two criteria to discard unreliable clusters from a binary brain map: 1. Cluster size: removes clusters smaller than CL_SIZE_THRES voxels. 2. Neighbour density: within surviving clusters, removes voxels with fewer than NN_THRES neighbours (eliminates isolated or spikey voxels).

Usage

newpmap = fmri_cleanclusters(pmap, CL_SIZE_THRES, NN_THRES)
newpmap = fmri_cleanclusters(pmap, CL_SIZE_THRES, NN_THRES, CONN)

Inputs

PMAP: Thresholded binary brain map, vectorized (228453 x 1) or volume (91 x 109 x 91).
CL_SIZE_THRES: Minimum cluster size in voxels. Clusters smaller than this are removed entirely. Use fmri_csthreshold to obtain a statistically motivated value.
NN_THRES: Minimum number of neighbours a voxel must have to be retained (range 0-26). Default: 0 (disabled).
CONN: Connectivity criterion for spm_bwlabel cluster detection: 6 = face connectivity (most conservative) 18 = face + edge connectivity (default) 26 = face + edge + corner connectivity (most liberal) Default: 18.

Outputs

NEWPMAP: Cleaned binary map, same format as PMAP input.

Examples

clean = fmri_cleanclusters(thresh_map, 50, 0);
clean = fmri_cleanclusters(thresh_map, cs, 2);
clean = fmri_cleanclusters(thresh_map, 50, 0, 6);   % strict face-only
clean = fmri_cleanclusters(thresh_map, 50, 0, 26);  % liberal corner