EEG BCI

Question 1

Name two examples where BCI is applied in a clinical setting and describe how it is applied.

Answer 1

1. For severly paralyzed patients (e.g. with ALS): BCI’s can be used to controll communication devices, media applications or protheses/wheelchairs
2. After stroke BCIs are used for faster rehabilitation in BCI guided rehab therapy

Question 2

What’s the simplest definition of BCI processing you can come up with?

Answer 2

BCIs do not “read minds” rather identify specific well defined signals and utilize them for further action

Question 3

Name three different (electrophysiological) methods of data acquisition that BCIs can be built upon.

Answer 3

1. Elecetroencephalography (EEG)
2. Electrocorticography (ECoG)
3. Intracranial microarrays of electrodes

Question 4

Based on EEG, name two features of measurement, the BCI systems can utilize

Answer 4

1. Recognition of changes in spontaneous oscillations
2. Recognition of event related potential

Question 5

In an EEG-setup where, would the hand movements be located.

NOT REALLY RELEVANT FOR EXAM, JUST FOR UNDERSTANDING

Answer 5

As the Hand movement is related to the motor and somatosensory areas of the contralateral cortex, the movements of the right hand should be found around the left central sulcus of the left hemisphere (C3) and vice versa (C4)

Question 6

What is the brain rhythm called, that is associated with the state of sensorimotor areas?

How does such rhythm behave when movements are performed?

Answer 6

The Sensory-motor rhythm (SMR).
The SMR´s power is rather high, when a muscle is idle, and is suppressed (power gets progressively smaller) when a movement is performed.

Question 7

How can the Sensory-motor rhythm be utilized in the scope of a BCI design?

Answer 7

As the modulation of SMR is similar for performed and imagined actions, imagining movement of the left hand will lead to a decrease of the SMR power over the right motor cortex (electrode C4) and vice versa.

When the person is idle, the power is greater as when the person imagines a movement.

The BCI is trained to recognize the different “power-states” over the C3/C4 and moves a cursor in the respective decision (dichotom decision variable).

Question 8

Name a basic setup that allows for manipulation of the P300 ERP component for later use in a BCI system.

Answer 8

You could use a classical Oddball-paradigm task, where participants attend to target stimuli with an enanced P3 amplitude as compared to standard stimuli. The BCIs are then based on recognizing differences in amplitudes.

Question 9

Name two BCI-systems that are used for selecting letters in a speller and explain how they work.

Answer 9

Matrix speller: The letters (Rows and columns) flash randomly. Flashing of target letter will elicit an enhanced P3 amplitude, whilst other letters evoke much smaller P30 amplitudes. BCI recognizes Amplitude and selects letter. (requires gaze-controll)

Hex-o-Spell: Similar design, using sets of letters that can be selected and letters within that subset. (works with impaired gaze-control)

Question 10

Name an example of a spelling BCI for the auditory modality.

Answer 10

Based on spatial auditory attention.

With a surround sound system (aluminium ring with speakers attached) different auditory can be presented from different directions. Subjects learn to attend to a certain position and a P300 is elicited as a result of the target stimuli.

Question 11

Name 5 potential challenges to the funcitoning of BCIs

Answer 11

1. Issue of EEG-volume conduction
2. subject-to-subject variability
3. session-to-session variability
4. trial-to-trial variablity
5. BCI Illiteracy

Question 12

Explain the issue of volume conduction in EEG.

Answer 12

The issue with volume conduction lies within the fact, that the brain tissue is generally a good conductor and that activity from e.g. deep neural generators is spread throughout the volume, resulting in a similar signal intensity at different locations of the scalp.

Consequently, if we have many sources, the different signals each arrive the different locations of the scalp.

Question 13

What is understood under “spatial smearing”

Answer 13

Spatial smearing in EEG (electroencephalography) refers to the phenomenon where electrical activity recorded from one area of the brain appears to spread or “smear” to neighboring electrode locations.

This results in high correlation of signals across the scalp

Basically the same idea as volume conduction!

Question 14

What is a current working hypothesis explaining BCI Illiteracy?

Answer 14

There is a possibility, that in the cortices of some patients the neural-generators of motor-movement is burried deep down in a sulcus, leading to a reduced readability of the signal. This again results in impaired BCI application.

Question 15

Explain the Linear Model of EEG.

Answer 15

First, it is assumed, that a single electrode may be affected by many neural generators and a single neural generator may affect multiple electrodes.

The linear model of EEG (LME) consists of two parts, that describe the relationship between the activity of the brain and the EEG signal: the forward model and the backward model.

In the forward model the recorded signal consists of the true signal and noise. The true signal is modulated through the mixing matrix (A), which describes the propagation of the signal from the source to the sensor.

The backward model assumes that an approximation of the true signal s is possible by applying an unmixing Matrix W to the recorded signal. Technically there is an infinite number of ways how this unmixing may happen, posing the Inverse problem of EEG, namely that a solution can not be unique.

Thus models impose certain constraints on the procedure of “unmixing”, e.g. by reducing the dimensionality of data as it is done in the principal component analysis

Question 16

Name and describe a prototypical procedure to classify data that is used in BCIs.

Answer 16

The linear discriminant analysis (LDA)

The LDA uses the mean of each class of data as well as the noise (which is represented through the covariance of classes) in order to assess, whether a new data entry is more likely to belong to a certain class or another.

Time for some fun mathematics! (:

Assumptions: data of each group is normally distributed with a certain mean and the same variance

The LDA is given by a vector W, that represented the difference of the means multiplied by the inverse covariance of the group. Graphically this vector points towards the mean of the group a new entry belongs to.

Furthermore a Bias “b” describes the transpostion of the vector W with th mean of the means, providing a line, that splits the groups.

Question 17

Explain how the machine learning approach works in BCI applications.

Answer 17

With an EEG-device:

In the calibration phase, the continuous EEG data is split up into feature vectors representing the data in its respective time points. These feature vectors are labelled in accordance with their belongings to a task or condition.

The LDA is trained with this data, a vector W and a bias B are acquired.

In the feedback application the vector is applied in a linear function (w T x + b) to a new data entry.

T is the symbol for a matrix transposition

Question 18

In 2016 the paper “The point of no return in vetoing self-initiated movements” by Schultze-Kraft et al. was published.

Briefly summarize the topic & hypothesis of the study, the methods, and the key results.

Answer 18

Topic & Hypothesis: The paper wants to address the questions if a movement can be cancelled after the readiness potential, and if so, if there is a point of no return, where a movement can not be cancelled anymore.

Methods: Participants played a duel game against a computer, where they had to press their foot down, when the screen was green, if they performed the action when the screen was red they lost and if they did nothing and the screen was red, they had a draw.

In phase one a BCI was calibrated to track, whether the RP was present, in stages 2 and three the BCI was activated when the RP was recognised and the screen turned red.

In stages 2 and 3 silent trials were performed, where the stop signal was recorded, but no red screen was shown.

Results: In silent trials, the BCI was able to predict button presses…
1. Movement can be cancelled after RP onset
2. There is a point of no return at around 200 ms before movement onset

Hier nochmal drüber, ist ja ganz schön lau die Karte