help with 2 samples permutation test on source adata

💬 Support & Discussions
1
  • MNE version: 1.8.0
  • operating system: Windows 11

I used sLORETA to obtain STC files for two groups of subjects under the same conditions (left and right brain)
I want to use the " ‘2 samples permutation test on source data with spatio temporal clustering’" to compare the results of two groups of participants and identify clusters with significant differences
then i use code below to show the Clusters with significant differences

        stc_all_cluster_vis = summarize_clusters_stc(
            clu, tstep=tstep, vertices=fsave_vertices, subject=subject
        )
        brain = stc_all_cluster_vis.plot(            
            subject,
            hemi="both",
            views="lateral",
            subjects_dir=subjects_dir,
            time_label="时间范围 (ms)",
            clim=dict(kind="value", lims=[0, 1, 40]),

However, the plot give me a strange result(shown in picture): half brain has no significant differences, half brain has all significant differences.

0c6e1cbcd5437a12f913803696b9938
0c6e1cbcd5437a12f913803696b99381230×1153 182 KB

there must be something wrong with my code, but there is no Reference Example to study,
Few people use MNE for 2-sample permutation tests on source data on google scholar.

Can anyone provide corresponding tutorials (preferably published literature)? :smiling_face_with_tear:

My complete code is as follows

def run_cluster_analysis(X1, X2, src, subjects_dir, subject='fsaverage'):
    """
    Perform spatiotemporal cluster analysis
    
    Parameters:
    X1, X2 : numpy.ndarray
        Group data, shape (n_subjects, n_vertices, n_times)
    src : SourceSpaces
        Source space object
    subjects_dir, subject : str
        Subject information
    """
    n_subjects1, n_vertices, n_times = X1.shape
    n_subjects2 = X2.shape[0]
    
    # KEY STEP 1: Transpose dimensions for cluster test
    # From (n_subjects, n_vertices, n_times) to (n_subjects, n_times, n_vertices)
    X1 = np.transpose(X1, [0, 2, 1])
    X2 = np.transpose(X2, [0, 2, 1])
    X = [X1, X2]
    
    # KEY STEP 2: Calculate spatial adjacency matrix
    adjacency = spatial_src_adjacency(src)
    
    # KEY STEP 3: Set statistical threshold
    p_threshold = 0.05
    f_threshold = stats.distributions.f.ppf(
        1.0 - p_threshold / 2.0, n_subjects1 - 1, n_subjects2 - 1
    )
    
    # KEY STEP 4: Perform cluster analysis
    F_obs, clusters, cluster_p_values, H0 = clu = spatio_temporal_cluster_test(
        X,
        adjacency=adjacency,
        n_permutations=1000,
        threshold=f_threshold
    )
    
    # KEY STEP 5: Select significant clusters (p < 0.05)
    good_cluster_inds = np.where(cluster_p_values < 0.05)[0]
    
    if len(good_cluster_inds) > 0:
        # KEY STEP 6: Prepare visualization data
        tstep = 1000 / 50  # Assuming 50Hz sampling rate
        fsave_vertices = [s['vertno'] for s in src]
        stc_all_cluster_vis = summarize_clusters_stc(
            clu, tstep=tstep, vertices=fsave_vertices, subject=subject
        )
        
        # Visualization code (simplified)
        brain = stc_all_cluster_vis.plot(
            subject, 
            hemi="both",
            subjects_dir=subjects_dir,
            time_label="Time range (ms)",
            clim=dict(kind="value", lims=[0, 1, 40])
        )
        
        return brain, stc_all_cluster_vis
    else:
        return None, None


def main():
    """Main function - simplified for clarity"""
    # Setup and data loading (pseudocode)
    env = setup_environment()
    X1 = load_group_data(group1_subjects, env['group1_dir'])
    X2 = load_group_data(group2_subjects, env['group2_dir'])
    
    # Run the analysis with loaded data
    brain, clusters_stc = run_cluster_analysis(
        X1, X2, env['src'], env['subjects_dir'], env['subject']
    )

`