EEG BCI

Question 1

What is BCI?

Answer 1

Brain-Computer interface!

A system that measures brain activity and uses decoded information for various purposes.

(E.g. Paralyzed people can communicate using EEG-data and smart spelling)

Question 2

BCI systems can’t read your mind but they can:

Answer 2

Recognize specific well defined control signals with a certain probability for some people.

Speed, accuracy and robustness is still limited

Question 3

Three ways to measure electrical brain activity:

Answer 3

EEG (electrodes on scalp)
ECoG (electrodes on top of cortex)
Intracranial microarrays (Electrodes deep in tissue, INVASIVE AF)

Question 4

Which electrodes are specifically important in EEG for BCI?

Answer 4

The electrodes placed above the central sulcus, as they represent signals from sensory and motor cortex, which is often used in BCI.

Question 5

Neural Activity: Spontaneous Oscillations

Answer 5

Oscillations occurring in specific parts of the brain as a response to action/inaction.

E.g. sensorimotor rhythm (SMR) which is a rhythm observed in the sensorimotor area. Here, the amplitude of SMR decreases when participants activate the body part represented by the corresponding brain area (Or even just imagines activating the body part).

Similarly, Visual Alpha Rhythm is used to describe the oscillations in the occipital cortex when eyes are open/closed. Power of signal decreases when eyes are open.

Question 6

Neural Activity: Transient Activity

Answer 6

Event-related potentials (ERP). Activity time-locked to an event (stimulus) and can be modulated by attention.

Question 7

Imagine the power spectrum of EEG, and explain the 1/f law

f=frequency

Answer 7

The 1/f law means that generally in the brain, as the frequency increases, the power decreases. Generally, low frequency bands have high power and high frequency bands have low power.

There are exceptions to this though, namely Alpha, Beta and Gamma waves

Question 8

Topographic Mapping of sensorimotor cortex

Answer 8

Mapping is contralateral

Hands represented approximately below electrode C3/C4

Question 9

Idle rhythms

Answer 9

Rhythms that are attenuated during activation. When neurons have nothing to do they synchronise and produce a stronger signal

E.g SMR and Visual Alpha Rhythm

Question 10

Whats going on here?

Answer 10

Paralysed participant is imagining moving either right or left hand. The EEG data is immediately analysed to determine whether there is left or right activation, and this is used to move a cursor in the desired direction.

Question 11

How does the P300 based BCI work?

Answer 11

With EEG, a P300 effect is shown when a target is attended to. Thereby, the BCI is controlled by increasing the attention to one stimuli while ignoring others.

This can e.g. be used in a Matrix Speller where rows of letters flash and letters can be selected using the P300

Question 12

What problem arises due to differences in tissue conductivity of skull and cortex?

Answer 12

Spatial Smearing!

Because the conductivity in the cortex is very good, and the conductivity in the skull is very bad, electrodes on the scalp will pick up electrical signal from many different parts of the cortex, causing the signal in every EEG channel to be highly correlated.

Think of it as the electrical signal from a specific part of cortex being smeared all over the skull (nasty)

Question 13

What’s the problem of subject-to-subject variability?

Answer 13

While there are general patterns to search for in certain activations, there is large subject-to-subject variability, which reduces the generalisability of BCIs. Often times, BCI’s have to be trained on subject-specific data (time consuming and inefficient)

Question 14

What is the Linear Model of EEG used for?

Answer 14

A model for tidying up messy EEG data in an attempt to grasp the activity at the actual source of activation.

The Model provides a forward model, which describes how the signals of sources in the cortex propagate to the electrodes on the scalp. Thus, the output of the forward model is a signal describing the linear relationship between the sources of activation + noise.

Thereafter, a backward model is used to de-mix the signals, so that the output reflects the activation at the source. The backward is typically some form of ML-model, such as LDA or PCA.

Question 15

Independent Component Analysis (ICA)

Answer 15

ICA is used to extract components that are maximally independent. Often times this is used to remove irrelevant components and artefacts in the data such as eye blinks, non-functioning electrodes or muscle tension.

Question 16

Linear Discriminant Analysis (LDA)

Answer 16

ML-model for separating two or more classes. Used when looking for specific patterns in the EEG data, e.g. thinking of moving right or left hand.

Question 17

Which data do you need to train a classifier for imagined left/right hand movement?

Answer 17

Labeled data is need, which is obtained by scanning the participant and instructing them when to imagine right or left hand movement at certain times. This data can be divided into two parts:
EEG signal of imagining right hand movement
EEG signal of imagining left hand movement

Different features from this data is fed to a classification algorithm, which is then trained and can be used to predict new data.

Question 18

How was BCI for readiness potential (RP) used to study free will?

Answer 18

Participants duelled against the computer, and they had to press a pedal with their foot if a light on the display was green. If they press it while green they get a point, if they press when red, they lose a point, if they miss the green light, it’s a draw.
A BCI was trained to identify RP in order to change color of the light when a RP is detected.

They examined whether participants were able to abort their movements after the RP, and showed that they were in some cases, primarily when the signal changed shortly after RP. Similarly, they compared the RP to the timing of Event-Related Desynchronization, which is an EEG measure of voluntary movements.

They showed that early cancellation of movement is possible even after RP, however there is a point of no return at 200 ms before movement onset where it is too late to cancel your movement.

Question 19

Are people able to influence their Readiness Potential(RP) and how was this studied?

Answer 19

No

People were asked to do button presses and got feedback which showed the strength of the RP signal. They reliably measures the RP strength, but across 300 trials, participants were not able to significantly reduce the RP.