Hello everyone,
I am trying to calculate Event-related changes of EEG activity such as event-related desynchronization (ERD) or event-related synchronization (ERS) like G. Pfurtscheller. First I want to show you how I processed my Data in Matlab:
1. First I segmented my Data based on the EMG Stimuli. (Then I had 14 Trials with 1000 Samples = 8s, because my sampling frequency is 125 (The Trials consist 4s of Rest-Periods and 4s of movement Periods).
2. I applied a Bandpass Butterworth filter on the Trials:
fs = 125;
[c,d] = butter(2,[8 30]/(fs/2),'bandpass');
for i = 1:14
C3_filt_Trial{i} = filtfilt(c,d,Segmente_C3_roh(:,i));
end
3. Then I calculated the Power for all Trials by squaring:
for i = 1:14
P_event{i} = C3_filt_Trial{1,i}.^2;
end
4. After that I averaged the Power for all Trials, I got 1000 Samples for P_event_average:
for i = 1:14
P_event_average = mean(P_event{1,i},2);
end
5. And then I calculated the Power for the Rest Periods:
for i = 1:14
P_rest_squared{i} = (C3_filt_Trial{1,i}(1:500,:)).^2;
end
6. The Power for the Rest Periods was then averaged, so I had 500 samples for P_rest_average:
for i = 1:14
P_rest_average = mean(P_rest_squared{1,i},2);
end
7. Finnaly I want to use the Equation:
ERD/ERS = (P_event_average)-(P_rest_avg)/(P_rest_avg)*100%
The Problem is that the arrays have different dimensions, due to the rest period which consists 0-4s of the whole trial. I am unsure if my processing pipeline in general is correct. I would really appreciate your help.
Greetings
Hi Faruk,
The power from the rest period should also be averaged over samples, giving you one representative power value for the rest period. Then, you will be able to apply your equation for each point of your trial, allowing you to observe the power variations during your whole trial relative to your rest period.
Some other points:
- I think it is more cautious to take a narrower window for the rest period, notably as some power changes may occur even before the movement, because of the anticipation of it.
- 8-30Hz seems pretty wide, and you may be interested in investigating more specific frequency bands. If it is the case, one way to do it is to bandpass filter your signal a bit wider (1-50Hz for example), and then apply a time-frequency transformation method (wavelet for example) to extract power for each frequency over time. Then, you will still be able to average your values over time (to compute your rest period) but also over the frequencies that you are interested in (maybe you will want to dissociate alpha-band (~8-12Hz) and beta-band (~13-30Hz) for example). You will also be able to plot time-frequency maps that will help you to have a better overview of your data.
I hope it is useful,
Good luck!
PM
- Badges
- Science topic