[Mne_analysis] spatio_temporal_cluster_1samp_test

[Emma Chen]

Dear MNE experts,

I have a question about creating a mask for
spatio_temporal_cluster_1samp_test. I'm trying to restrict the cluster
analysis on the occipital and temporal region for both left and right
hemispheres. The way I tried is as following:


   1. ###########  Exclude_indicies except for occipital and temporal ######
   2. label_lh = mne.read_labels_from_annot('fsaverage',
   parc='PALS_B12_Lobes', subjects_dir=subjects_dir,hemi='lh')
   3. label_rh = mne.read_labels_from_annot('fsaverage',
   parc='PALS_B12_Lobes', subjects_dir=subjects_dir,hemi='rh')
   4.
   5. stc_Ocp_lh = stc.in_label(label_lh[3])
   6. stc_Tmp_lh = stc.in_label(label_lh[5])
   7. stc_Ocp_rh = stc.in_label(label_rh[6])
   8. stc_Tmp_rh = stc.in_label(label_rh[8])
   9.
   10. Ocp_vertices_lh = stc_Ocp_lh.vertices[0]
   11. Tmp_vertices_lh = stc_Tmp_lh.vertices[0]
   12. All_vertices_lh = stc.vertices[0]
   13. Ocp_vertices_rh = stc_Ocp_rh.vertices[1]
   14. Tmp_vertices_rh = stc_Tmp_rh.vertices[1]
   15. All_vertices_rh = stc.vertices[1]
   16.
   17. Ocp_indices_lh = np.searchsorted(All_vertices_lh, Ocp_vertices_lh)
   18. Tmp_indices_lh = np.searchsorted(All_vertices_lh, Tmp_vertices_lh)
   19. Ocp_indices_rh = np.searchsorted(All_vertices_rh, Ocp_vertices_rh) +
   2562
   20. Tmp_indices_rh = np.searchsorted(All_vertices_rh, Tmp_vertices_rh) +
   2562
   21.
   22. temp_all_indices = range(5124)
   23. Exclude_indices_1 = np.setdiff1d(temp_all_indices, Ocp_indices_lh)
   24. Exclude_indices_2 = np.setdiff1d(Exclude_indices_1, Tmp_indices_lh)
   25. Exclude_indices_3 = np.setdiff1d(Exclude_indices_2, Ocp_indices_rh)
   26. *Exclude_indices* = np.setdiff1d(Exclude_indices_3, Tmp_indices_rh)
   27.
   28. ###########  Compute statistic ###########
   29. print('Computing connectivity.')
   30. connectivity = spatial_tris_connectivity(grade_to_tris(4))
   31.
   32. #    Note that X needs to be a multi-dimensional array of shape
   33. #    samples (subjects) x time x space, so we permute dimensions
   34. X = np.transpose(X, [2, 1, 0])  # X.shape was [5124, 81, 9],
   vertices * times * subjects
   35.
   36. #    Now let's actually do the clustering. This can take a long
   time...
   37. p_threshold = 0.05
   38. t_threshold = -stats.distributions.t.ppf(p_threshold / 2.,
   n_subjects - 1)
   39. print('Clustering.')
   40. T_obs, clusters, cluster_p_values, H0 = clu = \
   41.     spatio_temporal_cluster_1samp_test(X, connectivity=connectivity,
   n_jobs=2,
   42.                                        spatial_exclude =
   *Exclude_indices*,
   43.                                        threshold=t_threshold)
   44. #    Now select the clusters that are sig. at p < 0.05 (note that
   this value
   45. #    is multiple-comparisons corrected).
   46. good_cluster_inds = np.where(cluster_p_values < 0.05)[0]
   47.
   48. ###########  Visualize the clusters ###########
   49. print('Visualizing clusters.')
   50. fsave_vertices = [np.arange(2562), np.arange(2562)]
   51.
   52. #    Now let's build a convenient representation of each cluster,
   where each
   53. #    cluster becomes a "time point" in the SourceEstimate
   54. stc_all_cluster_vis = summarize_clusters_stc(clu, tstep=tstep,
   55.                                              vertices=fsave_vertices,
   56.                                              subject='fsaverage')
   57.
   58. #    Let's actually plot the first "time point" in the
   SourceEstimate, which
   59. #    shows all the clusters, weighted by duration
   60.
   61. # blue blobs are for condition A < condition B, red for A > B
   62. stc_all_cluster_vis.plot(subject='fsaverage', hemi='both',
   views='lateral',
   63.                                  subjects_dir=subjects_dir,
   time_viewer = False,
   64.                                  time_label='Duration significant
   (ms)', colormap = 'mne')


-----------End of the code---------

I thought I selected the vertices of occipital and temporal lobes
using *spatial_exclude
= Exclude_indices. * However, when I visualize the clusters, there are
still clusters show up in other regions e.g. frontal lobe, although overall
there are less clusters as expected when compared to when using
*spatial_exclude
= None.*

It's likely that the way I define Exclude_indices is wrong. It would be
greatly appreciated if anyone can help me correct this.

Thanks in advance,
Emma

------
Emma(Wei) Chen, Ph.D.
Objects and Knowledge Laboratory
New York University Abu Dhabi
PO Box 129188
Abu Dhabi, United Arab Emirates
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20170124/07849ec1/attachment.html