Understanding how Epochs work

💬 Support & Discussions
1

I have annotated EEG data (raw; continuous) spanning 900 s. The start of Event A coincides with t = 0 and runs for 30 s. Followed by Event B which runs for the next 30 s. This cycle repeats till the end of the data.

So, I execute the following code for epoching into the default 0.7s segments:

raw = ...
epochs = mne.epochs(raw, mne.events_from_annotations(raw)[0], event_id=mne.events_from_annotations(raw)[1], preload=True);
Ev_A_epochs = epochs['Event A']
Ev_B_epochs = epochs['Event B']

So, 15 epochs corresponding to Event A and another 15 corresponding to Event B are created.

Query: All epochs (taken together) span (15+15) * 0.7 = 21 s which means 900 - 21 = 879 s worth of data was lost. Am I correct? Such being the case, how were these 0.7s segments sampled from the 30s events (almost 43 times as long) during the creation of epochs?

2

The tmin, tmax are relative to the event onset. For instance, if you have a continuous recording of 10 seconds with a trigger onset at t=1 second, then tmin=-0.5, tmax=2 will create an epoch which starts at t=0.5 (event - 0.5) and ends at t=3 (event + 2). It’s simply cropping the data.

import numpy as np

from matplotlib import pyplot as plt
from mne import create_info, Epochs
from mne.io import RawArray


info = create_info(["AF7"], 1024, "eeg")
data = np.random.randn(1, 5120)
data[0, 800:1200] *= 3
raw = RawArray(data, info)
events = np.array([[1024, 0, 1]])  # event at t=1 second (1024 samples)
epochs = Epochs(
    raw, events, event_id=dict(my_evt=1), baseline=None, tmin=-0.5, tmax=2
)

raw_data = raw.get_data()  # (1, 10240) - (1 channel, n_samples)
epo_data = epochs.get_data()  # (1, 1, 2561) - (1 epoch, 1 channel, n_samples)

f, ax = plt.subplots(2, 1, sharex=True)
ax[0].plot(raw.times, raw_data[0, :])
ax[0].set_title("Raw data")
ax[1].plot(epochs.times + 1, epo_data[0, 0, :])  # event is at t=1 second
ax[1].set_title("Epoch positionned on the same time-axis")

image

The default 700 ms window is good in most ERP analyses, where you have an event that elicits a short brain response. In this case, you consider the 200 ms before the event comes up as a baseline and you study the reaction to the event in the following 500 ms.

2 Likes