1 of 45

Working with EEG Data

Module 4:

CS 198-96: Introduction to Neurotechnology

1

2 of 45

MODULE 4 OVERVIEW

3

What is EEG?

1

Data Analysis in Python

2

EEG Signal Processing

Working with EEG Data

2

3 of 45

WHAT IS EEG?:

EEG INTRODUCTION

MODULE 4: WORKING WITH EEG DATA

3

4 of 45

SECTION OVERVIEW

3

Introduction

1

4

Some EEG Terminology

Types of EEG Data Analysis

2

Brain Waves

4

5 of 45

WHAT WE KNOW...

Neurons communicate using both electrical and chemical signals.

Action potentials are all-or-none electrical signals carried along neurons.

Postsynaptic potentials are changes in the membrane potential of the postsynaptic terminal of a chemical synapse.

5

6 of 45

What is EEG and what can it detect?

What are two properties that EEG signals should have in order to easily detect brain activity?

What are advantages and disadvantages of using EEG?

KEY QUESTIONS:

6

7 of 45

Measures electrical activity of the brain

Detects activity of large groups of active neurons at the same time

Primarily measures postsynaptic potentials (not action potentials)

Has a variety of clinical and research applications

WHAT IS AN EEG?

7

So as a reminder, EEG is placed on a person’s scalp to detect the electrical activity of neurons in the cerebral cortex. It can be used to measure brain activity that occurs during an event or to measure spontaneous brain activity. It detects the signals created when populations of neurons fire at the same time. EEG primarily measures postsynaptic potentials, or changes in membrane potential that are elicited by neurotransmitters binding to receptors on the postsynaptic membrane. Action potentials in the axons do not contribute much to the ongoing EEG activity as they are too short, they go in too many directions relative to the surface of the cortex, and they are not synchronized. The roots of the EEG lie in the work done by English physiologist Richard Caton (Key-ton) in 1875. He made electrical recordings form the surface of dog and rabbit brains using a primitive device that is sensitive to voltage. The human EEG was first described by Austrian psychiatrist Hans Berger in 1929 who observed that waking and sleeping EEGs are distinctly different. Today, the EEG is used primarily to help diagnose certain neurological conditions, especially the seizures of epilepsy, and for research purposes, notably to study the stages of sleep and cognitive processes during wakefulness.

8 of 45

Detected electrical signal represents the summed extracellular electrical field potentials generated by EPSP and IPSP on dendrites and neuronal cell bodies of the pyramidal neurons.
An EPSP in a dendrite leaves a negative charge in the extracellular space surrounding the synapse. A dipole is therefore generated in which the EEG can detect.

PHYSIOLOGICAL BASIS OF EEG

8

9 of 45

DETECTING BRAIN ACTIVITY

For brain electrical activity to be detectable through the skull, there must be a strong signal summed over many neurons that are:

All behaving similarly at the same time
All oriented in the same way

So that negative and positive don’t cancel each other out when summed.

The pyramidal cells in the cortex have the right properties (see image).

9

Signals cannot be detected from single neurons because the potentials are far from the scalp and have a very small magnitude. The signal detected is the summation of all the dipoles created at each of the thousand neurons. In order for the brain activity to be easily detected, the signals that are summed must have two main properties. The first property is that the neurons must behave similarly. In other words, a group of neurons must be activated altogether so that the same charges are propagated at the same time. The second property is that they must be oriented in the same direction. If these two properties aren’t met, the charges cancel each other out, and so the electrical activity isn’t detected. The pyramidal neuron in the cortex, which is shown here, is a great example for satisfying these properties.

10 of 45

High temporal resolution

well-suited for fast, dynamic, temporally sequenced cognitive events (cognitive, perceptual, linguistic, emotional, motor, etc)

Directly measures neural activity

oscillations observed in EEG signal are direct reflections of neural oscillations in the cortex

EEG signal is multidimensional (see figure)

time, space, frequency, and power & phase

WHY EEG? (ADVANTAGES)

10

There are several reasons why high-temporal resolution techniques such as EEG are exceptional tools for studying neurocognitive processes.

The first reason is that these methods capture cognitive dynamics in the time frame in which cognition occurs. Cognitive, perceptual, linguistic, emotional, and motor processes are very fast. Most cognitive processes occur within tens to hundreds of milliseconds. Furthermore, cognitive events occur in a temporal sequence that may span hundreds of milliseconds to a few seconds. High-temporal-resolution techniques are well suited to capture these fast, dynamic, and temporally sequenced cognitive events. By comparison, the temporal precision of the hemodynamic response obtained in fMRI, for example, is 2-3 orders of magnitude slower than that of the electrophysiological response. This is why most brain computer interfaces, including prosthetics, and many other neurotechnologies out there, take advantage of EEG’s high temporal resolution.

A second reason why tools such as EEG are advantageous for studying neurocognitive processes is that they directly measure neural activity. The voltage fluctuations that are measured by EEG are direct reflections of biophysical phenomena at the level of populations of neurons. Furthermore, oscillations that can be observed in the EEG signal are direct reflections of neural oscillations in the cortex. This can be seen as an advantage over currently used MRI-based functional measures, which do not measure neural events directly and have a more complex relationship between what they actually measure and what can be inferred in terms of the kinds of neural dynamics that produce or correlate with a hemodynamic response.

A third advantage of EEG is that the EEG signal is multidimensional. Although you might conceptualize EEG data as two-dimensional or basically, voltage changes over time and space where space is measured through different electrodes, in fact, EEG data consists of at least four dimensions: time, space, frequency, and power and phase. We’ll discuss what they are later on in this section. For now, you should know that this multidimensionality provides many possibilities for specifying and testing hypotheses that are rooted both in neurophysiology and in psychology. The brain can be conceptualized as an enormously complex biological system that uses a multidimensional space for information processing, representation, and transfer.

11 of 45

Poor spatial resolution

not well suited for problems that require precise functional localization

Not well suited for deep brain structures
Suboptimal for studies concerning cognitive processes that are slow and have an uncertain and variable time course

fMRI is better suited to studying slow cognitive processes

WHY NOT EEG? (DISADVANTAGES)

11

Of course, EEG is not perfect. EEG is a powerful and insightful brain-imaging technique, but it is not well suited for addressing all research questions or creating types of neurotechnology that require high spatial resolution. EEG, due to its poor spatial resolution, is not well suited for scenarios in which precise functional localization is important. If your research problem involves testing whether different striatal subregions are preferentially involved in two different cognitive processes, EEG will be of little use to your research. However, if your research problem requires dissociating activity in parietal cortex from activity in frontal cortex, the spatial accuracy of EEG is likely to be sufficient to confirm functional or anatomical dissociations at this scale. Nevertheless, EEG is likely to be a suboptimal methods for any research questions involving “where in the brain does process X occur or is information Y stored.” Another EEG disadvantage is that EEG is not well suited for testing hypotheses about deep brain structures, although activity from deep brain structures can be measured with EEG in SOME cases which we won’t have time to cover. EEG may also be a suboptimal brain-imaging technique for questions concerning cognitive processes that are slow and that have an uncertain and variable time course. That sort of defeats the purpose of using EEG because we mostly want EEG for its high temporal resolution. The extremely high temporal precision of EEG might be disadvantages in some cases. Many EEG analyses become unreliable when the cross-trial jitters in timing are longer than a few tens or hundreds of milliseconds. The relatively low temporal precision of fMRI is better suited to studying slow cognitive processes. Some social and emotional processes might have slow and uncertain time courses that can be more difficult to interpret with EEG.

12 of 45

Summary/What’s Next

Summary:

EEG can be used to directly measure brain activity during an event, or to measure spontaneous brain activity.

What’s next:

Brain Waves

AUTHORS: CZARINAH MICAH RODRIGUEZ

12

13 of 45

WHAT IS EEG?:

BRAIN WAVES

MODULE 4: WORKING WITH EEG DATA

13

14 of 45

SECTION OVERVIEW

3

Introduction

1

4

Some EEG Terminology

Types of EEG Data Analysis

2

Brain Waves

14

15 of 45

WHAT WE KNOW...

EEG can be used to measure spontaneous brain activity or brain activity that occurs during an event.

For brain electrical activity to be detectable through the skull, there must be a strong signal summed over many neurons that behave similarly and orient the same way at the same time.

15

16 of 45

What are Neural Oscillations?

What are the different types of brain waves?

What do brain waves tell us about different states of consciousness?

KEY QUESTIONS:

16

17 of 45

Also known as brain waves
Repetitive patterns of neural activity in the central nervous system
Visualized in terms of frequency and measured in Hertz (one cycle per second)

NEURAL OSCILLATIONS

17

18 of 45

Any signal can be decomposed into multiple sinusoidal waves.
Each sinusoidal wave (cos, sin) oscillates at a different frequency and the algorithm extracts which frequencies are more prevalent in a signal.

FOURIER TRANSFORM

18

19 of 45

BRAIN WAVES OVERVIEW

19

Bold = cursor point

Let’s first explore what EEG can measure by looking at patterns of brain activity associated with particular states of consciousness or behavioral states. We’ll be talking more about these specific brain waves in the next few slides in more detail.

As shown in figure A [A], when a person is aroused, excited, or even just alert, the EEG pattern has low amplitude or wave height, and high frequency so there are a higher number of brain waves per second. This beta rhythm is typical of an EEG taken anywhere on the skull of an alert participant, not only in humans but other animals as well. In contrast, [B] when a person is calm and resting quietly, especially with the eyes closed, rhythmical, larger, slower brain waves emerge as shown in figure B. These so called alpha waves wax and wane in amplitude at a frequency of approximately 11 cycles per second. In humans, the largest alpha rhythms come from the region of the visual cortex at the back of the head. If a relaxed person is disturbed or opens his or her eyes, the alpha rhythm stops abruptly. Not everyone displays alpha rhythms, and some people display them much more consistently than others.

EEG is a sensitive indicator of states other than arousal and relaxation. Figures C through E illustrates the EEG changes that take place as a person moves from drowsiness [C] to sleep [D] and finally enters deep sleep [E]. As the EEG rhythms become slower in frequency and larger in amplitude, theta waves and then delta waves [E] are produced.

These distinctive brain-wave patterns make the EEG a reliable tool for monitoring waking and other conscious states. Slower waves also occur during anesthesia. Brain trauma may also result in EEG slowing, [F] making EEG useful for evaluating how severe a head injury is like in this figure which shows a comatosed eeg signal. If the brain ceases to function or there’s brain death, the EEG trace becomes a flat line.

20 of 45

DELTA WAVES (1-4 Hz)

Slow-wave sleep
Stronger delta waves = deeper sleep

20

21 of 45

THETA WAVES (4-8 Hz)

Memory encoding, retrieval, mental arithmetic
Drowsiness

21

Theta waves are 4 to 8 Hertz. Two types of theta oscillations have been described based on the location of their occurence. The hippocampal theta rhythm is a strong oscillation that can be observed in the hippocampus, while the cortical theta rhythms are low-frequency components of scalp EEG. There is no clear understanding of the function of theta waves, however a combination of several theories supports that hippocampal theta waves are linked to navigation and memory. Theta waves are known to be present during cognitive tasks such as memory encoding and retrieval, and mental arithmetic. Theta waves increase in power with memory demands. Meanwhile, Cortical theta waves are observed more frequently in young children and appears during meditative, drowsy,hypnotic, and non-deep sleeping states.

22 of 45

ALPHA WAVES (7.5-12.5 Hz)

Attention and awareness
Relaxed Wakefulness

22

23 of 45

BETA WAVES (13-30Hz)

Concentration and attention
Planning and execution of movements

23

24 of 45

GAMMA WAVES (30-70 Hz)

Consciousness and perception
Sensory processing and information uptake

24

25 of 45

Indicators of neurological phenomena such as:

Sleep and the state of consciousness
Motor control
Perception and information processing
Pattern Generation
Memory
Abnormal neural function, such as epilepsy, and Parkinsons

WHY DO NEURAL OSCILLATIONS MATTER?

25

26 of 45

THE 10-20 SYSTEM

26

27 of 45

ELECTRODE MONTAGE

27

28 of 45

Time
Space
Frequency
Power and phase

EEG’S FOUR DIMENSIONS

28

As mentioned before, EEG signal is multidimensional and it consists of at least four dimensions: time, space, frequency, and power and phase. Time is often measured in seconds and frequency is measured in Hertz which is the number of cycles per second. Power is the strength of the frequency-band-specific activity, and phase is the timing of the activity. Power and phase are discrete elements of a dimension because they provide largely independent information. As you can see form this graph, the x axis indicates time in seconds and the y-axis indicates frequency in hertz. Power is shown in color where red indicates the strongest frequency band specific activity. Space is another dimension that can be mapped using the international 10-20 system electrodes which you can see here in the bottom.

29 of 45

THE FREQUENCY SLICE

29

30 of 45

THE TIME SLICE

30

31 of 45

THE SPACE SLICE

31

32 of 45

THE TIME-FREQUENCY SLICE

32

33 of 45

Rhythmic - waves of approximately constant frequency

Arrhythmic - no stable rhythms are present

Dysrhythmic - rarely seen in healthy subjects

FREQUENCY PROPERTIES

33

34 of 45

Synchrony - the simultaneous occurrence of EEG waves over distinct regions on the same or opposite sides of the head
Asynchrony - non-coherent occurrence of EEG activities over regions on the same or opposite sides (hemispheres) of the head.
Hypersynchrony - when describing EEG patterns that are attributed to increased synchronization of neuronal activity

Hypnagogic hypersynchrony - occurs at the onset of sleep in normal infants and children

SYNCHRONY

34

35 of 45

Monomorphic - distinct EEG activity appearing to be composed of one dominant activity
Polymorphic - distinct EEG activity composed of multiple frequencies that combine to form a complex waveform.
Sinusoidal - waves resembling sine waves. Monomorphic activity is often sinusoidal.
Transient - an isolated wave or pattern that is distinctly different from background activity.

Spikes and sharp waves

WAVEFORM MORPHOLOGY

35

The shape of a wave or an EEG pattern is determined by the frequencies that combine to make up the waveform and by their phase and voltage relationships. Wave patterns can be described as being monomorphic, polymorphic, sinusoidal (saynusoydal), and transient. Monomorphic waveforms involve distinct EEG activity appearing to be composed of one dominant activity. In contrast, polymorphic waveforms consist of distinct EEG activity composed of multiple frequencies that combine to form a complex waveform. Sinusoidal waveforms have waves resembling sine waves, and monomorphic activity is often sinusoidal. Lastly, a transient waveform is an isolated wave or pattern that is distinctly different from background activity. Examples include spikes which is a transient with a pointed peak and a duration from 20 to under 70 millisecond, and sharp waves which is a transient with a pointed peak and duration of 70-200 millisecond.

36 of 45

Summary/What’s Next

Summary:

EEG data consists of four dimensions: time, space, frequency, and power and phase. Each of these dimensions have several different properties.

What’s next:

The main types of EEG data analyses.

AUTHORS: CZARINAH MICAH RODRIGUEZ

36

37 of 45

WHAT IS EEG?:

TYPES OF EEG DATA ANALYSIS

MODULE 4: WORKING WITH EEG DATA

37

38 of 45

SECTION OVERVIEW

3

Introduction

1

4

Some EEG Terminology

Types of EEG Data Analysis

2

Brain Waves

38

39 of 45

WHAT WE KNOW...

EEG data comprises of four dimensions: time, space, frequency, and power and phase.

Each dimension has specific properties that can provide us with valuable information.

39

40 of 45

What is time-domain analyses?

What is frequency-domain analyses?

What is time-frequency domain analyses?

KEY QUESTIONS:

40

41 of 45

Event-Related Potentials (ERPs) - an electrophysiological response to a stimulus

TIME-DOMAIN ANALYSES

41

When it comes to time-domain analyses, we usually think of event-related potentials. An event-related potential or ERP is the measured brain response that is the direct result of a specific sensory, cognitive, or motor event. More formally, it is any stereotyped electrophysiological response to a stimulus. ERPs are not the only analyses that are possible in the time domain. There’s also micro states and multiscale dynamics or fractal based dynamics, but ERPs are the dominant form of analyses in the time domain. Here you see a bunch of single trial EEG data. The number here corresponds to the trial number and the x-axis indicates time in milliseconds. Time 0 is when the stimulus appears in front of the subject. So here you can see that individual trials can be variable so there’s a bit of variability over different trials. In this plot on the right, you can see all of the 99 individual trials in gray. And the black part here is plotting the ERP which is the average of all these trials. To compute an ERP, you take all of the trial voltage fluctuations at each time point and average them together. So just sum and divide by N and that gives you the ERP. When looking at this plot, the ERP is kind of smaller in magnitude compared to single trial fluctuations in gray. ERPs tend to be an order of magnitude smaller than the voltage potentual fluctuations of the individual trials. You can get very different interpretations of this. One interpretation is that single trial data is very noisy. The large fluctuations may be indicative of a lot of brain noise or measurement noise. So by averaging over the trials, we can get rid of the noise and focus on the relevant parts of the signal. Alternatively, these fluctuations can be interpreted as important signals that could be relevant for understanding cognitive processes and you can lose them in the time domain averaging. The graph at the bottom sort of shows the same ERP but more zoomed in so you can see it in finer detail.

42 of 45

FREQUENCY-DOMAIN ANALYSES

42

The next type of EEG data analysis is frequency-domain analyses. The idea of frequency-domain based analyses is to represent and analyze the signal not in the time domain but in the frequency domain. So on the top left you can see the time domain and here’s a one second tick mark so this is two seconds in total. You can see that this signal has three peaks so this is three hertz as shown here at the bottom for the amplitude vs frequency plot. So figure A2 is the representation of this signal in figure A1 and figure B2 is the representation of the signal in figure B1 and so forth. You can see that in figure B1 is oscillating much faster and has a lower amplitude of around 8 Hertz so that’s why the amplitude of the frequency here is lower compared to figure A. And finally, figure C is the sum of the signals from figures A and B. You can see that the constituent parts, the components of the signal in figure C1, is much easier to identify in the frequency domain in figure C2 compared to the time domain in figure C1. This advantage of the frequency domain is much more prominent in this signal on this upper right figure. This looks like a very complex and noisy signal so it’s difficult to identify how many different components are in this signal in the time domain but in the frequency domain it’s much easier. We can see that there’s around 1, 2, 3, 4 sine wave components of this signal plus some low amplitude broadband noise in here. So it’s easier to talk about this signal in terms of frequency domain. You will learn how to get to the frequency domain in the next section using a famous signal processing method called Fourier transform.

43 of 45

FREQUENCY-DOMAIN ANALYSES

Fourier Transform

43

One of the main disadvantages of the fourier transform is that it doesn’t show us the temporal information. All of the temporal information is actually in the fourier transform BUT in the phases which are not generally shown. But from the results of the fourier transform, it’s difficult to determine exactly when different frequency components are present. So in this plot, we can say that there’s such and such hertz component in this signal, say like 8 hertz or something. But from this plot with only the frequency domain, it is not possible to determine whether the 8 hertz component is constant over time in the EEG signal or whether there is a very strong 8 hertz in the very beginning or whether it appears in short bursts and so forth. We can get none of these temporal dynamics in these plots, but in fact these temporal dynamics are things that we should be very interested in because the brain is very dynamic. It is highly non-stationary. A solution to this is doing morlet wavelet convolution so instead of getting the dot product of the EEG data with a sine wave like in fourier transform, we do the dot product of the EEG data with a morlet wavelet, which I won’t discuss any further since it’s out of scope but feel free to look more into wavelet convolution if you’re interested.

44 of 45

Morlet Wavelet Convolution:

Tradeoff: temporal and frequency precision

TIME-FREQUENCY-DOMAIN ANALYSES

44

But that part about trying to obtain both temporal and frequency precision is what time-frequency-domain analyses is all about. Unfortunately there’s a tradeoff between temporal dynamics and frequency precision, which I won’t go into detail how but in general, if you try to get high temporal precision, you lose some frequency precision, and vice versa. But anyways, with time-frequency-domain analyses, we can have plots such as this where you can see time on the x-axis and frequency on the y-axis with smaller numbers corresponding to slower oscillations and higher frequency numbers corresponding to faster oscillations. And the color or the depth in the plot corresponds to a variety of features that we can obtain using a time frequency based approach. In this particular example, power is plotted and here power is converted to a decibel change from baseline. So power refers to the energy in the signal and so red means relatively higher amplitude oscillations and blue would mean relatively lower amplitude oscillations relative to the pre stimulus baseline.

45 of 45

Summary/What’s Next

Summary:

The main types of EEG data analyses include time-domain analyses, frequency-domain analyses, and time-frequency-domain analyses.

What’s next:

EEG Signal Processing

AUTHORS: CZARINAH MICAH RODRIGUEZ

45