The best way to extract neural oscillations is to perform a Fourier transform on your preprocessed data and then plot the resulting frequency patterns in the category of brain waves your interested in seeing. You can further preprocess your data to exclude certain channels, or target specific frequency ranges to observe features of the neural oscillation. See NeuroTechX.edu “Preprocessing” for a detailed look at the preprocessing steps that can be applied to your data.
Before extracting neural oscillations there are several steps that must be undertaken to prepare you data:
In the below example I have used a dataset created by experimental runs by (research reference) so when the data is fetched it will have already underwent some preprocessing which will not be covered in either examples. However in both cases the data will be fetched using the below commands:
Line 2 is relevant in this example, as there are 14 experimental runs to choose from that were performed in this study and each was tested under different conditions. In this experiment run 3 measured the EEG signal obtained during movement of the left and right hands, both separately and simultaneously.
When plotting power spectral density (psd), only epoching is necessary as we want to see the psd across the entire available frequency range. However for topomap plotting you will need to preprocess you data with a band pass filter to isolate the specific frequency range you want to visualize. The example below also strips the channel names of their default “.” keys to avoid errors when reading your channel names.
Epoching is basically segmenting your data into smaller chunks of readable data that contain events, or fluctuations in the EEG signal (caused by changing signal potential?). Epoching helps to truncate the objects you need to analyze into more manageable bytes that contain the information you actually want. To epoch your data you can follow the steps below:
The above defines the events you want to look for, as well as the type of signal your program should expect to analyze (meg vs. eeg vs. eog). I selected the eeg=True as the data we will be looking at was obtained using an EEG headset. Below we continue the necessary inputs for epoching:
The above code defines the epochs which will be further manipulated. Try the function plot.epochs() to see how your data has transformed! (Hint: It might look something like the image below!)
The following steps will diverge depending on the methods of visualization you want to apply. Below I will highlight how to use psd to visualize dominant frequencies as well as topomaps to highlight specific frequencies in localized areas.
Power spectral density measure the power of a signal in (UNITS). The function epochs[events].plot_psd() can be used to plot specific epoched events as a function of power spectral density over a specific frequency range. Below is an example of a script you can run to achieve this. Note that the dataset imported from eegbci is already preprocessed and transformed using the Fourier function, so the steps you should see below are:
Figure 1: Power spectral density plotted for the electrode C3 located in the left side of the motor cortex. It is evident that the psd content is most evident in the alpha range and theta range, evident of neural oscillations occurring during movement of the right hand.
Figure 2: Power spectral density plotted for the electrode C4 located in the left side of the motor cortex. It is evident that the psd content is most evident in the alpha range and theta range, evident of neural oscillations occurring during movement of the left hand.
A classifier will be applied to visualize the presence of neural oscillations using a binary discrimination to highlight whether you have the presence of waves at certain frequencies or not. While there is a bit of variability on visualizing the strength of your signal, this is really simply a “there” or “not there” method of looking at neurons firing over specific electrodes. Understanding which electrodes correspond to which regions of the brain will help you to understand where the majority of activity is occurring in the brain.
Plotting PSD to visualize frequency is limited to specific channels and may require excess work to view the local presence of waves. By contrast, topomap offers specific feedback on the local presence of waves. The classifier assembled in the below example uses a binary range characterized by the red positive values and the blue negative values. The total range is from -0.1 to 1.0 and refers to the electric potential with 1.0 indicating the presence of a measurable signal.
There are several more steps involved with topomap visualization. The first occurs in the preprocessing step as you must apply a band pass filter (raw.filter(fmin,fmax)) to isolate the frequency range that interests you. Refer to the above section where I break down each wave into their respective frequency range. The below example will be using a band pass filter for alpha and beta waves since the eegbci dataset runs are based on on motor cortex activity.
Five arbitrary time points were chosen in the below example to highlight the shift in wave presence over time. Since the run involves squeezing the right and left hands, we should expect an increase in the alpha wave presence (specifically the mu frequency) and a reciprocal decrease in beta wave frequency in the area corresponding to either right or left hand (opposite sides of the brain from the hand involved). Timepoints 4 and 5 illustrate this difference:
Figure 3: Topomap of alpha waves (7.5-12.5 Hz) during movement.
Figure 4: Topomap of beta waves (13-30 Hz) during movement.