cluster-based permutation tests on spatial data

💬 Support & Discussions
1

Dear MNE community,

I am using permutation_cluster_1samp_test() to compare two conditions:

t_obs, clusters, cluster_p_values, H0 = permutation_cluster_1samp_test(diff,
out_type = ‘indices’,
n_permutations = 10000,
adjacency = adjacency,
threshold = threshold_tfce,
tail = 1,
seed = 77)

with the adjacency matrix computed from an EEG layout file:

adjacency, ch_names = find_ch_adjacency(epochs.info, ch_type=‘eeg’)

I have 400 observations of the difference between the two conditions per electrode. So the shape of my data is (400 x 64) and is only spatial (on the sensor-level). My question is whether this is the correct approach so that clusters are formed spatially based on the adjacency matrix I defined (and my second dimension not interpreted as a time-series)?
Also, by using TFCE am I rendering the adjacency matrix in a sense obsolete because each point is tested for significance independently? How is max_step used with TFCE?

Best,
Chantal

2

hi @coderb

yes to me it should be correct but I don’t use these often. In case of doubt I would do a tiny simulation personally.

Also, by using TFCE am I rendering the adjacency matrix in a sense obsolete because each point is tested for significance independently?

TFCE is using clustering so the adjacency should be used.

How is max_step used with TFCE?

TFCE is restarted multiple times on the remaining insignificant electrodes.

HTH
Alex

3

@agramfort thank you for the reply and the clarifications! I will try out a simulation just to be sure.