PSD topomap plot of 4 different classes

Support & Discussions
,
1

Hello all,

I have an motor imagery eeg dataset which has four classes: left hand, right hand, both feet and tongue. I am trying to plot topomaps of power spectral density of each class. I am just wondering that what brain region activated during each class so I have plotted psd topomaps. I know that during motor imagery the alpha band should be considered. But when I look for plots, there is no any discrimination between classes. I was expecting that during left hand motor imagery, the right side of the brain region should be activated, during right hand motor imagery, the left side of the brain region should be activated, during feet motor imagery, the center of the brain region should be activated, and so on. But the plots does not belong what I was expected (I am considering alpha band). I trust the data because the classification performances for this data is obtained high. The plots are shown below. Why this situtation happens? Thank you.

image
image912×725 181 KB

2

Did you consider a suitable time interval? Usually, discrimination is highest between 1 to 3 seconds after the cue. Furthermore, you might need to normalize to a baseline to see any differences.

2 Likes
3

Yes, I have tried many time intervals, and also baseline removal, but the plots are not what I expected. Maybe I was expecting wrong thing. For left hand motor imagery, I know the right side of the brain should be activated and the power increases. However, I read now that a decrease in alpha power over the primary motor cortex region(right side of the brain) corresponding to the left hand area is expected. I am not sure which one is correct. The last results after applying baseline removal and selecting the time interval between 1 and 3 is shown below. Thanks for your helping.
The code for obtaining epochs is shown below:

events = mne.find_events(raw)

event_id = {"left_hand":1, "right_hand":2, "both_feet":3, "tongue":4}

tmin, tmax = 0, 4

epochs = mne.Epochs(raw, events, event_id, tmin, tmax, proj=True, picks=['eeg'],baseline=(0,1), preload=True)

image
image594×719 110 KB

4

Motor imagery is associated with contralateral power decrease relative to baseline. For now, I would focus on channels C3, Cz, and C4 and look at time/frequency maps as described in this example. Using this as a template, you should be able to see the expected patterns (if they are present in the data and your preprocessing is correct).

4 Likes