Lecture 3: Basics of digital signal processing & Neuronal oscillations: Mechanisms, diversity and functions?

Question 1

What do oscillations reflect?

Answer 1

A resonance between neurons that cause synchrony to emerge.

Question 2

What is depicted in these pictures?

Answer 2

Pyramidal cells and interneurons are mutually connected. On the picture on the right you see that pyramidal cells (purple) fire just before interneurons (red). So what happens is that:

• Pyramidal neurons fire and connect to interneurons.
• Interneurons are excited and connect to pyramidal cells.
• This causes inhibition of pyramidal neurons.
• When the inhibition wears off, pyramidal neurons can fire again, starting all over again. This gives the local field potential (LFP) a rhytmic structure.

Question 3

What are local field potentials (LFP) and how are LFPs different from EEG signals?

Answer 3

LFP is the electric potential recorded in the extracellular space in brain tissue, typiclly using micro-electrodes. They differ from EEGs, because EEGs are recorded on the surface of the scalp, and with macro-electrodes.

Question 4

What is the difference between endogenous and exogenous firing of neurons?

Answer 4

Endogenous is spontaneous (or self-organized) while exogenous is stimulus induced.

Question 5

Fill in:

Complex …-scale dynamics emerge from … interactions.

Answer 5

Complex large-scale dynamics emerge from local interactions.

Question 6

A model with 2500 intergrate-and-fire neurons are arrenged in a 2D grid. Here, 75% is excitatory and 25% of the neurons are inhibitory. What ‘phenomenon’ in this model contributes to the the statement that complex large-scale dynamics emerge from local interactions?

Answer 6

The fact that neurons only connect within their local range. In this way you can research the influence of connectivity on brain activity. You can for example tell the computer to only connect 25% of the neurons. What effect does 25% connectivity have on brain activity?

Question 7

What does this picture depict?

Answer 7

This picture depicts how good we have gotten at estimating e.g. alpha-frequency bands/oscillations and the excitatory and inhibitory ratio. The bottom picture displays EEG oscillations that are then measured by MEG. Here, the red displays alpha oscillations that are measured by MEG. The picture above, displays the model computed by an algorithm. Here you can see that based on the amplitudes, the model can predict how large the oscillation will be. So a high peak is equal to a lot of excitation, while a small peak is equal to no excitation and thus inhibition.

Question 8

What is qualitative EEG?

Answer 8

The visual inspection of patterns, which is the clinical standard (requires up to 5 years of training). For example: you can recognize seizures based on local and global patterns.

Question 9

Because doctors can’t monitor a patients’ EEG for 24 hours, there needs to be a solution for qualitative EEG. What is quantitative EEG (qEEG)?

Answer 9

Digital signal processing algorithms and statistics, based on comparison between brain areas.

Question 10

qEEG can be divided into classical and modern qEEG. What is the difference?

Answer 10

• Classical qEEG: measures power spectra (how strong/frequent are certain signals) and coherence (synchrony/correlation between two brain areas).
• Modern qEEG: analysis of time-frequency, phase-locking, time-series, etc.

Question 11

Both the classical and the modern qEEG analysis approaches can be divided into two main categories. What two?

Answer 11

1. Characterization of fluctuations of individual traces (local).
2. Characterization of correlations in the fluctuations of traces recorded over spatially distinct brain regions.

Question 12

What is the sampling theorem (Nyquist or Shannon)?

Answer 12

A continuous signal can be completely recovered from its samples if, and only if, the sampling rate is greater than twice the highest frequency of the signal.

Question 13

What is aliasing?

Answer 13

It occurs if the sampling theorem is violated: the original signal cannot be reconstructed from the digitized signal.

So as you can see in the picture, the highest frequency is around 21 while the sampling frequency is 23. The sampling theorem only holds when the sampling rate is greater than twice the highest frequency of the signal. And of course this doesn’t apply to this signal.

Question 14

• In what type of person are delta waves mostly found?
• In what type of person are theta waves mostly found?
• Where in the brain is alpha mostly measured?
• Where in the brain is beta mostly measured?

Answer 14

• Delta, measured in infants and sleeping adults.
• Theta, measured in children and sleeping adults.
• Alpha, measured occipitally (sensory).
• Beta, measured frontally and parietally.

Question 15

Sort the different types of frequencies from highest to lowest based on their amount of Hz that they display.

Answer 15

• Gamma >30 Hz
• Beta, 13-30 Hz
• Sensori-motor rhythm, 12-18 Hz
• Alpha, 8-13 Hz
• Theta, 4-8 Hz
• Delta, 0.5-4 Hz

Question 16

What is a power or amplitude spectrum analysis (in classical qEEG)?

Answer 16

Like with the color white, you can decompose the EEG signal into waves with different frequencies and strength. This is based on the Fourier transformation.

Question 17

When deconstructing a signal into waves of different frequencies, why is the sum of component waves not enough to reconstruct the original signal? Why does this result in a deviation of the original signal, as you can see in the picture?

Answer 17

Because when deconstructing the signal into three different waves of different frequencies, it seems as the ratio between the three different waves is 3. This is actually not the case, so you have to know the ratios between the different frequencies.

Question 18

What can you compute with this formula (Fourier transformation) if you give a signal and put this signal in this formula?

Answer 18

You can compute the relative intensity of different frequency components that are derived from this signal in a so-called power spectrum.

Question 19

Match oscillations (delta, thelta, alpha, sensori-motor rhythm (SMR), beta, gamma) with a type of activity (relaxed, attention, drowsiness, deep sleep, physical rest, problem solving).

Answer 19

• Delta, deep sleep
• Theta, drowsiness
• Alpha, relaxed
• SMR, physical rest
• Beta, attention
• Gamma, problem solving

Question 20

What is:

• the common goal of power spectra?
• the typical application of power spectra?

Answer 20

• Common goal is to quantify frequency distribution of fluctuations in neuronal signals.
• The typical application is measuring ongoing neuronal activity in two experiments (healthy vs. patients, rest vs task)

Question 21

There are two types of power spectra: absolute vs normalized. What is the difference?

Answer 21

• Absolute power spectra tells you the power of the entire signal in microvolts, it cannot make a distinction between the different frequencies.
• Normalized (or relative) spectra, define the entire spectrum from e.g. 1-30 Hz to be 100%. Now, each frequency band may be quantified relative to the rest.

Question 22

What are disadvantages of using the absolute power spectrum?

Answer 22

Subjects may differ 10-fold in absolute amplitude (e.g. due to skull conductivity). So a task-related change of 10% of one subject can correspond to a 100% change in the other subject. Thus with the use of absolute power spectra, their is bias by reactivity of subjects with high amplitude.

Question 23

What are disadvantages of normalized power spectra?

Answer 23

The effect is not isolated to one frequency band, an increase in one frequency band will lead to a decrease of other frequency bands (e.g. delta goes up from 20% to 30%, other frequency bands will then have a lower percentage (total of 10%).

Question 24

What is the working memory?

Answer 24

The process by which the brain sustains the activity of cells whose firing ‘represents information’ dervied either from brief sensory input or readout from long-term memory.

Question 25

What is the spatio-temporal code and why does this prove that oscillations are an important link between a sensory input or read from long-term memory and execution of this input?

Answer 25

The spatio-temporal code says that oscillations provide a temporal reference for a destributed representation of information in neuronal assemblies. Thus when oscillations can't be generated (e.g. when interneurons are inhibited), excitatory pyramidal cells can still fire. But the signal isn't picked up any further (no synchronization), which results in he absence of execution of this input. This was studied in a landmark study. It provides empirical support for a causal role of oscillations in information processing and a link between two important levels of neuronal organization: **single-cell firing** and the **formation of synchronous cell-assemblies in functional networks**.

Question 26

To study the role of brain oscillations in working memory, the scalp EEG during the retention interval of a modified Sternberg task was recorded. This way, differences in brain activity between remembering letters or forgetting them (working memory) have been measured. What was the outcome?

Answer 26

Power spectral analysis revealed a clear peak at 9-12 Hz --\> alpha band frequency (8-13 Hz). This alpha peak systematically increased with the number of items held in working memory. (This was found to be in contradiction with previous ideas that stated that alpha band oscillations represented brain idling.)

Question 27

What type of oscillation was measured during a working memory task and what was the origin of the signal?

Answer 27

Theta peak, a radial dipolar source in the cortex (remember that de electrode located the dipole in the cortex, due to its orientation --\> see picture).

Question 28

This picture depicts the evoked potentials that results from cell assemblies passing on their short activity to neigboring cell assemblies. It now seems that this process is evoked or induced by transient oscillations. Thus the dichotomy between ERPs and oscillations is disappearing. So we need a way to analyze these transient oscillations. For this time frequency power representation is used. Why do we use this and not a normal power spectrum?

Answer 28

A power spectrum reflects oscillations (bottom picture), here you can see that there's a strong alpha peak. But what you don't see is temporal resolution, like you can see in the upper picture. So a time frequency power representation can be made by making a time frame of 1 second. Here you put the time on the x-axis and the frequency on the y-axis. This way, you can clearly see that there's alpha activity that starts at 0.2 seconds.

Question 29

Considering this picture, we see that if an oscillation is evoked by a stimulus or task, but not very well time/phase locked to the stimulus onset, then the average evoked potential will not reveal the the oscillation is present. What do we do the reveal the presence of oscillations?

Answer 29

We perform a time-frequency analysis. We will then come to the conclusion that there's gamma oscillation evoked by the stimulus, but also gamma oscillations a few seconds later when the participant notices/acts upon the stimulus. But since this reaction is not time-locked, you would not see this induced gamma oscillations in the TF-plot of the averaged ERP.

Question 30

The global workspace theory states that in order to gain conciousness, simultaneous activation of brain regions is needed. By stimulating the medial nerve at threshold level (so that sometimes participants were aware of the stimulus and sometimes not), you can research whether and when conciousness arises. What was the conclusion?

Answer 30

You can clearly see in the picture that when participants were not aware of the stimulus, alpha band activity wasn't present. This tells us that alpha oscillations in widespread brain regions is a hallmark of consciously perceiving the weak somatic stimuli.

Question 31

What is special about delta oscillations?

Answer 31

Unlike other oscillations that arise from excitatory pyramidal cells and inhibitory interneurons, delta oscillations arise from cells that initiate a burst of activity which is passed on to other neighboring cells, which then receive a lot of input, that again pass on this input to other neighboring cells (exponential activation). Delta oscillations seem to emerge at different locations and travel around the cortex at night.

Question 32

How do we know delta oscillations are important for learning and memory?

Answer 32

An experiment where the delta oscillations during the deep sleep phase (stage 4) were gently (so that the participants wouldn't wake up from it) interrupted. The morning after MRI studies then revealed that participants activated their hippocampus less during certain tasks and also could not learn as good in comparison to control.