Brain Computer Interfaces explained in 10 minutes
A BCI can be divided into three subtypes
1. Active BCI
The input is directly controlled by the user, just for the purpose of controlling something
2. Reactive BCI
The input is indirectly controlled by the user. The reaction to an external stimulus controlls something
3. Passive BCI
the input is taken from brain activity without the purpose of voluntary control.
In other words the Brain Signals just get used to enrich the human computer interaction with additional information
Source: Zander et al., 2009
In the classic setup an EEG is placed on the head of a subject. From there cables continue into a computer that contains the software and algorithms to analyze the data. Additionally an interface device is used to communicate with the subject.
Modern EEG setups usually contain a wireless EEG Headset that is connected with just one device that handles the processing and interface at once.
What devices can be used to pick up Brain Signals?
Of course the sources are our neurons. Basically you can imagine the neurons as a dipole, or in other words: like a battery.
The neuron can furthermore be divided into the dendrites, which are the “input arms” of the neuron. These are connected with other neurons and receive signals from them.
The middle of the cell is called “Somata” and contains the genetic information and the important cell organelles. In this Somata the neuron roughly said “decides” whether or not to send a signal down the axons if a signal is received from the dendrites.
The bottom part of a neuron is called axon and is the place where the cell “sends” output signals to other cells. As the dendrites the axons are connected to other cells and transmit information.
Now the problem for a Brain Computer Interface is that the charges that a single neuron produces are really really small (0.2 pA). Therefore only many neurons that fire synchronously in the same direction can be observed by an EEG or MEG. The smallest measurable size are 50.000 Neurons (10 nA) which corresponds to an area of 0.63 mm².
But one important thing is that those neurons have to fire in the same direction, otherwise the charges might cancel out themselves. For this purpose so called pyramidal cells are the perfect matches. They are oriented in the same direction and distributed around the grey matter in our brain.
But when do 50.000 neurons fire near-synchronously?
Event-Related Potentials (ERPs) and oscillatory processes are the two major BCI-detectable EEG and MEG phenomena.
There are roughly three activities where this happens:
An external event triggers a cascade of related neural processes
An internal event triggers a cascade of related neural processes (for example the typical “Aha!” effect)
Neural populations enter a synchronized steady-state pattern
Picking up the EEG Signal
The EEG has the problem that the signal travels from the neuron through different layers of the skull that have all different conductive properties. This results in a slight distortion of the signal and makes the localization of the source more difficult.
Furthermore the signal that arrives at the electrode is really weak and needs to be amplified (around 50.000 times).
Another problem is that most of the “raw” signal is noise. There are different types of noise, which can be distinguished as following:
Non Brain Artifacts
50/60 Hz line
sensor related (DC offset drifts, thermal noise,…)
e.g. Head movements
peak and rebound amplitude
mainly in frontal EEG recordings
Luckily the common noise gets filtered out during the amplification (analog filtering). But after that one of the main problems of doing a Brain Computer Interface starts: Filtering the noise out of the signal to be left with the data you are interested in. But we will come back to this point later.
Now after amplification and basic filtering we are left with around 128 different recordings of EEG data. But what now? We can look at the data at different time points, but this delivers us just some numbers that tell us nothing. To use the EEG data we need to reference it to another electrode to say something about the activity of the brain region.
To make this clear let us look at an example:
Imagine a town that is built on a river. You measure the height of your house (10 meters) and the church (50 meters). The next day the tide rises and the church is after measuring it again just 48 meters high and your house just 8 meters. Now pushed to the extreme, a storm flood comes in and the church is now just 10 meters high and your house -30. It would be way easier to compare their heights if you take the height of the church in reference to your house, which would be the difference in height between them: 40 meters. This number never changes and allows for a comparison.
The same is done with the EEG data. You take the measured result in comparison with another electrode. There are three common ways to do this:
The reference point is just the average of all electrodes. This is the most common used method today.
The reference point is one special electrode in the front center of the skull (top view), called FCZ. The result can be quite different depending on the distance of the recorded electrode to this reference point, and is therefore not always the best method to measure
The reference point is the average of two electrodes, similar to the FCZ. The electrodes used are located on both sides in the middle of the skull.
In general it is a fact that there exists no neutral reference. Therefore it is important to always keep an eye out for which reference point was used in a study, because results can greatly differ from method to method.
The Magnetoencephalography is quite similar to the EEG. It picks up magnetic signals from the brain, not electric currents like the EEG, and transforms it into data.
Therefore it uses a Pickup-Coil and a device called SQUID that measure different magnetic fields of the neurons. The neurons do not only produce electric current, they do produce additionally a magnetic field. The advantage of this magnetic field is that it is not as heavily distorted as the electric current by the layers of the skull. This allows a better localization of the source of the signal compared to the EEG.
The thing with this magnetic field is that it is really really weak and that it needs to be shielded from other magnetic fields (for example the earth magnetic field). This can be achieved with a faraday’s cage in which the measuring unit is placed, it basically shields the inside from all magnetic currents.
Inside this shield the above mentioned Pickup-Coil and the SQUID unit are placed. The Pickup Coil is a superconductive metal component that is placed in the matching liquid (in this case Liquid Helium) which enables the superconductive properties (in this case the metal reaches this state at -269°C).
There are three basic kinds of Pickup Coils:
The Magnetometer is only one ring. Even though it provides the highest amplitude of results measured, it has relatively strong measuring differences depending on the position over the head.
Two rings are placed on top of each other. This way neutral currents get canceled out, but the signal is not as strong as with the Magnetometer
Two rings are placed besides each other. Neutral currents cancel each other out again, but again there is less strength of the signal.