How to conduct the significant test on the decoding results?

Support & Discussions
1
  • MNE-Python version: 0.23.0
  • operating system: Linux

Hi! Guys,

I’m wondering how to conduct the significant test in decoding analysis.
Here is my code:

epo_data = epoch.get_data()
labels_ani = epoch.metadata['feature_animacy']
n_cvsplits = 10
cv = StratifiedKFold(n_splits=n_cvsplits)
clf = make_pipeline(
    StandardScaler(), # Z-score data, because gradiometers and magnetometers have different scales
    LogisticRegression(random_state=0, n_jobs=12, max_iter=500)
)
sliding_ani = SlidingEstimator(clf, n_jobs=12)
sliding_ani.fit(epo_data, labels_ani)
auc_fea_ani = cross_val_multiscore(sliding_ani, epo_data, labels_ani, cv=cv, n_jobs=12)
auc_fea_ani_avg = np.nanmean(auc_fea_ani, axis=0)

And here is the plot of the result:

image
image752×323 81.4 KB

I’m trying to figure out how to find the significant cluster in the time dimension, and produce the plot like this in which the significant time range was marked:

image
image1334×1094 144 KB
.

I searched in MNE’s tutorial, but there seems no topic about the significance test in decoding analysis, did I miss it? or are there some functions that can help me do the significance test?

Thanks a lot! :sparkling_heart: :sparkling_heart: :sparkling_heart:

2

I don’t know any code to do this type of plot.

but I am sure it would be interesting to MNE folks

can you give it a try? maybe find some code we could get some inspiration from?

Alex

3

Hi! Alex,
Actually, I’m more interested in how to find the significant cluster, and I think maybe I found a function that can be used to test the significance of decoding: mne.stats.permutation_cluster_test — MNE 0.23.0 documentation
Maybe I can feed the decoding time series obtained from the decoding process, and mne.stats.permutation_cluster_test() might return the statistical values of each time point. I’m not sure if it will work but I might give it a try, what do you think?

Best

4

yes sounds good

Alex