EEG Data and How to Interpret It

EEG data consists of time-series recordings of electrical activity from multiple electrodes placed on the scalp. Each electrode captures voltage fluctuations over time, resulting in a multi-channel dataset that reflects the brain's dynamic activity.

How to obtain EEG Data

EEG data can be easily obtained through various means: as of Fall, 2025, we're using Healthy Brain Network's (HBN) open EEG dataset (release 10) which can be found here. This dataset includes EEG recordings from a large number of participants, along with relevant metadata such as age, sex, and clinical assessments.

Doing this research, we're focused on classifying and predicting people's brain state given some EEG data. This is the entire goal of Digital Twins, which is to create a digital representation of an individual's brain state based on their EEG data. This can be useful for a variety of applications, such as diagnosing neurological disorders, monitoring cognitive states, and developing brain-computer interfaces.

When working with EEG data, it's important to preprocess the data to remove artifacts (e.g., eye blinks, muscle activity) and to filter the signals to focus on specific frequency bands of interest (e.g., delta, theta, alpha, beta, gamma). After preprocessing, various analysis techniques can be applied, such as time-frequency analysis, connectivity analysis, and machine learning algorithms for classification and prediction tasks.

Different Bands

EEG signals are typically analyzed in terms of different frequency bands, each associated with specific cognitive and physiological states. The main EEG frequency bands are:

Delta (0.5 - 4 Hz): Associated with deep sleep and restorative processes.
Theta (4 - 8 Hz): Linked to drowsiness, meditation, and light sleep.
Alpha (8 - 13 Hz): Related to relaxed wakefulness and closed-eye states.
Beta (13 - 30 Hz): Connected to active thinking, focus, and problem-solving.
Gamma (30 - 100 Hz): Involved in higher cognitive functions, such as perception and consciousness.

Useful Note

Anything relatively above 50 Hz is often considered noise and is usually filtered out during preprocessing (i.e. artificial electrical signals from muscle movement or external electronic devices).

Power Spectral Density (PSD)

Let's explore a fundamental part of EEG data analysis: Power Spectral Density (PSD) plots. PSD plots help us visualize how the power of the EEG signal is distributed across different frequency bands. By examining these plots, we can identify dominant frequencies and assess the overall brain activity.

Power Spectral Density is a measure of the power present in the EEG signal as a function of frequency. It provides insights into the strength of various brain wave components and helps in understanding the underlying neural dynamics.

PSD Example — Taken from https://still-breathing.net/tag/eeg/

As you can see from the above example, the x-axis represents frequency (in Hz), while the y-axis indicates power (in µV²/Hz). Peaks in the PSD plot correspond to dominant frequency bands, which can be linked to specific cognitive states or neurological conditions.

At the lower end of the frequency spectrum, we often observe higher power in the delta and theta bands, which are associated with sleep and relaxation. As we move towards higher frequencies, the alpha and beta bands become more prominent, reflecting alertness and cognitive engagement.

With these changes in power across different frequency bands, we can infer various aspects of brain function and identify potential abnormalities. For instance, an increase in theta power may indicate drowsiness, while elevated beta activity could suggest heightened cognitive processing.

Real Example

Let's use one of HBN's (release 10) EEG recordings to illustrate how to interpret EEG data. Below is a sample EEG signal from a participant during a resting-state session.

Resting State EEG — This is a sample resting-state EEG signal from HBN's dataset through a low-pass filter at 45 Hz to remove high-frequency noise. This is done in the notebook, `filter.ipynb`.

Above shows the processed signals from the channels (marked by E#) over time. Each line represents the voltage fluctuations recorded by a specific electrode on the scalp.

Luckily, HBN provides metadata for this resting state data, which includes instructions given to the participant during the recording. In this case, the participant went between durations of keeping their eyes open and closed. This information is crucial for interpreting the EEG data, as different brain states (eyes open vs. eyes closed) can significantly affect the recorded signals.

With some scripting (seen in separate_tasks_resting_state.ipynb), we can separate the EEG data into segments corresponding to eyes open and eyes closed periods. This allows us to analyze the differences in brain activity between these two states.

We can then plot the Power Spectral Density (PSD) for both conditions to visualize how brain activity varies with eyes open versus eyes closed.

PSD Eyes Open vs Closed — This is produced in the notebook, `test_alpha.ipynb`, showing the PSD comparison between eyes open and eyes closed conditions.

You can see, at around 10 Hz, there's a noticeable difference in power between the two conditions. The eyes-closed state typically shows a pronounced alpha peak (around 8-13 Hz), which is associated with relaxed wakefulness. In contrast, the eyes-open state often exhibits reduced alpha activity, reflecting increased sensory processing and alertness.

In Short

By analyzing EEG data through PSD plots, we can gain valuable insights into brain function and cognitive states. The differences observed between eyes open and eyes closed conditions highlight the dynamic nature of brain activity and its responsiveness to sensory input.

Summary

What we want to do in this research is to use these findings, and produce models able to predict and classify brain states based on EEG data. This in turn can lead to the development of Digital Twins that accurately reflect an individual's brain state in real-time.

Much of this is left vague on-purpose to allow for exploration and discovery as we progress through the research. As we delve deeper into EEG data analysis, we'll uncover more nuances and complexities that will inform our modeling efforts.