Stimuli and feature extraction methods for EEG-based BMIs: a systematic comparison

A brain-machine interface (BMI) can be generically defined as a communication
interface system between the central nervous system and an external device. As
a potential assistive technology to paralyzed or *locked-in* individuals
suffering from motor control disorders, a brain-machine interface (BMI) could
be designed to translate the brain activity into control commands for an
external device such as a computer cursor controller, an external actuator or a
prosthetic arm, to name a few. However, there are limitations to be overcome
before a successful implementation of a BMI for the control of an external
device, such as an exoskeleton or a prosthetic device, in a real-life
environment is achieved. BMIs are full control systems and they intrinsically
have diverse aspects affecting their performance rates, either directly or
indirectly. Non-invasive modalities can be adapted to be safely used outside
the experimental environment, but as the brain activity measurement is
indirect, the desired information is attenuated and distorted through the
tissues and becomes less differentiable from noise and other artifacts.
Classification accuracy above 70% is often considered in the literature as the
minimum performance requirement for a BMI for communication establishment and
user controllability. Multiple factors, including information transfer rate and
accuracy, must be considered in order to objectively compare the performance of
different systems. In spite of several attempts and initiatives from BMI groups
and members over the last two decades to provide standardization guidelines,
the comparison between studies and different BMIs remains a challenge.

The primary objectives are: 1) to systematically compare four brain stimulation
methods and different signal processing techniques and select the best method
for eventual development of a discrete brain-machine interface that allows the
translation of brain activity signals into two binary control parameters
(angular speed and position) for a single joint robotic arm, and 2) to select
the best sets of 8 and 32 EEG channels out of 64 channels based on accuracy
criteria, for portability increase and cost/setup time reduction purposes, by
evaluating the accuracy as a function of the number of channels.

Observational study
During the experiments, the brain activity will be registered using high
density 64-channel EEG integrated in a head cap. EMG and accelerometers from
different forearm and shank muscles will be recorded simultaneously for
movement detection and monitoring. A video camera will be recording the
participants during the experiments to allow movement identification and
verification. The participant will be seated in front of an LCD screen placed
approximately 60 cm away from his/her eyes. The participant will be instructed
to avoid excessive head, jaw and tongue movements to limit artifacts. After
experimental setup, the participant will be requested to perform tasks based on
different stimulation paradigms:
* Paradigm I: Steady State Visually Evoked Potential (SSVEP) task;
* Paradigm II: P300 task;
* Paradigm III: Motor Execution / Motor Imagery task;
* Paradigm IV: Hybrid task (SSVEP + Motor task, simultaneously).

Other parameters of interest will be collected from questionnaires. After
recording, the EEG and other sensors will be removed.

There are no risks or benefits, and the burden is limited to the time invested
in participation (approximately 180 minutes, including 60 minutes preparation
and set-up time).