A as ADHD and autism, and recentA as ADHD and autism, and recent

A Brain Computer Interface (BCI) is a system which allows someone to communicate
information about their mental state without the use of the peripheral nervous system. It does
so by detecting, decoding and rendering brain signals into an output that can later be used to
accomplish the user’s intent. There are many different uses and applications for such
systems: movement restoration, for example with a prosthetic hand, aid in communication
with the use of a visual speller or environmental control, such as learning to turn a switch on
or off through mental waves.
Other possible purposes are mental state monitoring, like workload monitoring or
awareness detection, enhancing gaming and entertainment experiences, and achieving a
change in perception or behavior (Brunner et al, 2015).
The last-mentioned goal is accomplished through neurofeedback, a technique that enables
individuals to change their brain activity by using an instrument that provides information on
cerebral functioning.
Neurofeedback works on the paradigm of operant conditioning, detecting and
rewarding “good” mental states to increase the likelihood that they’ll be expressed in the
future. It has been used in the clinical setting to treat disorders such as ADHD and autism,
and recent findings suggest that this might be a beneficial treatment for these individuals
(Kouijzer et al. 2011).
What are ERD/ERS?
Different types of events induce different changes in the power of the EEG signal.
In this experiment, we will consider the changes that are related to movement.
During a movement there is a modulation of power in mu (8-12 Hz) and beta (13-25 Hz)
bands that reflect a Desynchronization and a subsequent Synchronization of a population of
neurons in the brain, in this case the contralateral primary sensorimotor areas.
An Event Related Desynchronization (ERD) describes a decrease of oscillatory
activity in a given frequency band that is generated by an event compared to a baseline. An
Event Related Synchronization (ERS) instead expresses the opposite phenomenon. While
the former is a reliable correlate of the increased cellular excitability, the latter represents an
inhibitory state in the cell population.
Although the strongest effect are collected through actual movement, movement
imagery does give the same EEG signature as a physical action (even if the effects are
smaller) (Kuhn et al, 2005).
In between imagination and real activity, we can describe quasi-movements:
volitional undetectable movements that generate a stronger ERD than imagined ones
(Nikulin et al 2008).
In our research, we ask the subject to perform quasi-movements with their hand, to
produce a clearer EEG signal.
Aim of the research
The aim of this research is to determine whether time pressure can influence the
learning process of controlling a BCI and, if it does, in what way. The hypothesis guiding this
experimental design is that an increase in time pressure would result in a decrease in
learning ability, and a consequent decrease in performance and efficiency.
But why would we study this? BCI is a developing field in science and technology, it
still faces many challenges and it’s far from the sci-fi ideal of a psychic, intuitive system,
capable of instantly connecting with anyone’s brain without requiring any setup or calibration.
Low Signal to noise ratio, low spatial resolution and high levels of variability (both
inter subject and inter session) make building a usable BCI extremely hard.
One of the many goals of BCI research is to improve the efficiency of a system, by
increasing speed and accuracy.
By researching the effects of time pressure on the learning performance of controlling
a BCI, we could learn how the newly gained information can be used to increase the learning
performance of controlling a BCI. If the aforementioned hypothesis will be rejected, it
indicated that time pressure has little to no influence on the learning rate. This could be the
starting point to the development of new protocols to shorten the time required for training.
This will help people who currently refrain from learning to control a BCI due to the
required time to learn. Next to that it will help BCI users in general.
This report will first show what studies already have been done on the subject and
what similar research could be found in the literature. After that, the report will describe the
task to be executed, describing how time pressure is controlled in the different conditions
and how these conditions influenced the performance of the subject. This section also
includes the electrode placement, the performance evaluation before and after the different
training sections, the used pre-processing and the classification.
After this section the report will show the results.This data will be supported by EEG
graphs, frequency graphs and class confusion matrices.
Finally, we will review the results and draw the conclusions based on this study: if the
manipulation of perceived time pressure had an effect on the BCI learning rate. Furthermore
it will be discussed what further research can be done and what parts of our methodology
could be improved.
Even though EEG measurements only partially reflect the underlying neural
phenomena, these contain a significant amount of information on brain activity.
Unfortunately, the data also consists of a considerable amount of noise.
In order to extract the actual brain signal and discard confounding information from
the data we have collected, we need to pre-process it and ensure that it is fit for
classification.
There are many steps in the signal analysis procedure, the first one being
detrending. Detrending is the act of removing any linear trend in the data that may be
present.
The next stage is identifying bad channels and removing them, so they will not take part in
the classification stage and will not compromise the potential success of it.
Next up is data cleaning: that includes artifact detection and removal.
An important part of data cleaning is Signal filtering, that it is used in order to
attenuate background noise whilst leaving the signal intact, and hence make the target
signal easier to detect.
The filters we will use are Spatial filtering and Spectral filtering.
Spatial filtering, also called Re-referencing, consists of changing the EEG reference
location in order to remove the artifacts from the true signal. In this case, we will re-reference
using CAR (Common Average Reference).
Spectral Filtering involves applying a filter that cancels out certain frequencies, in this
experiment we will circumscribe the frequencies between 8 and 25 Hz to include the ? and
the ? rhythms .
The rhythm ?, which corresponds to an oscillation of signal EEG between the 8 and
13 Hz, is caught in the sensorimotor zone located in the central region. This rhythm shows
attenuation in its amplitude when some types of movement are performed, when the
intention is had to realize some movement, or simply imagining movements of the
extremities ( Roman-Gonzalez, 2012).
The rhythm ? , which corresponds to an oscillation of signal EEG between 18 and 25
Hz, is also modulated in many tasks. Typically, the ? activity is depressed during both
sensory and motor tasks. After the task, the ? activity “rebounds” (Hari and Salmelin, 1997).
The ? rhythm shows peak desynchronization over the vertex.
The last step is Bad trial Identification and removal.
For all of these steps we have used the provided code, as for our purposes this
preprocessing code is more than sufficient enough.