Sourcebased Brain Computer Interfaces
Lecturer

Petia Georgieva
University of Aveiro
Portugal


Brief Bio
Petia Georgieva (S’14) received the M.Sc. and PhD degrees in electrical engineering from Technical University of Sofia, Sofia, Bulgaria in 1989 and 2007, respectively. Since 2003 she has been a lecturer at the University of Aveiro, Portugal and researcher in the Institute of Electronics Engineering and Telematics of Aveiro (IEETA), Portugal. Dr. Georgieva was i) an invited professor in Rowan University, New Jersey in 2016; ii) a visiting faculty in Computer Science Department, Carnegie Mellon University (CMU), Pittsburgh, USA, 2012; iii) an invited researcher in Computer Science Department, University of Arkansas at Little Rock. USA, 2011 and iv) an invited researcher in the School of Computing and Communications, University of Lancaster, UK, 2011. Over the last ten years Dr. Georgieva works in the area of machine learning and data mining with strong focus on noninvasive Brain Computer Interfaces (BCI), brain neural activity recovering and affective neurocomputing based on EEG data. She has published more than 120 papers in peer reviewed journals and international conferences. Dr. Georgieva is a Senior member of IEEE, Senior Member of International Neural Network Society (INNS) and Elected Member of the Executive Committee of the European Neural Network Society (ENNS) for 20142016.

Abstract
During the last decade advances in many scientific fields have supported the idea that a direct interface between the human brain and an artificial system, called Brain Computer Interface (BCI), is a viable concept, although a significant research and development effort has to be conducted before these technologies enter routine use. The principal reason for the BCI research is the potential benefits to those with severe motor disabilities, such as brainstem stroke, amyotrophic lateral sclerosis or severe cerebral palsy. Moreover, recent advances in sensor technology and machine learning assert that using our brain for communication may have a significant impact in the way people will operate computers, wheelchairs, prostheses, robotic systems and other devices in the future. Among various alternatives, Electroencephalography (EEG)based BCI is the most studied noninvasive interface to build a direct communication pathway between the brain and an external device. However, correlated noises in EEG measurements still constitute a significant challenge. Alternatively, building BCIs based on filtered brain activity source signals instead of using their surface projections, obtained from the noisy EEG signals, is a promising new research direction. This tutorial will present the state of the art in sourcebased noninvasive BCI technologies. The talk will highlight research on spatialtemporal reconstruction of the underlying brain neural generators based on the EEG recording. Several source reconstruction approaches will be reviewed with a strong focus on the statistical statespace framework, where the source localization is formulated as the estimation of the posterior probability density function of the state based on available EEG observations.
Keywords
Brain computer interface, Electroencephalogram (EEG)
sourcebased BCI, Bayesian estimation, Particle filters, Beamforming
Aims and Learning Objectives
Introduction to BCIs, biosignal processing and data modeling
Target Audience
Young researchers focused into biomedical signal processing, feature extraction, patterns recognition, statistical signal processing
Prerequisite Knowledge of Audience
signal processing and filtering, linear algebra, Kalman filters, Bayesian estimation theory
Detailed Outline
The spatialtemporal reconstruction of the
underlying brain neural generators based on the EEG recording
has emerged as an active area of research over the last decade.
Several source reconstruction approaches, each employing a
different set of assumptions, have been proposed to overcome
this illposed inverse problem. They can be divided in two
main classes: i) imaging models (also known as current
density reconstruction models), which explain the data with
a dense set of current dipoles distributed at fixed locations;
and ii) equivalent current dipole models (also known as point
source or parametric models), which assume a small number
of focal sources at locations to be estimated from the data.
While the imaging techniques provide a detailed map of the
neuronal activity, the parametric models represent a direct
mapping from scalp topology to a small number of parameters.
Dipole solutions provide more intuitive interpretations that
explain the sensor data. Furthermore, it is easy to report statistics
of dipole parameters over different subjects. Summarizing
distributed brain activity with a small number of active dipoles
simplifies the analysis of connectivity among those sources.
Additionally, building BCIs based on the neuronal sources
instead of the EEG sensor data is gaining more interest.
In particular, sourcebased BCI seems an appealing alternative
to well known invasive solutions through implant placement
(intracortical electrodes) by neurosurgery.
Popular deterministic parametric solutions include the Multiple
Signal Classification (MUSIC) algorithm and its modified
versions, the methods for inverse problems,
the construction of spatial filters by dataindependent or
datadriven methods and blind source separation techniques.
However, these approaches are based on the
assumption that the brain source locations are known a priori
or perform a search of the overall head volume to find their
positions. Given the spatial source locations, they estimate the
amplitudes and directions of the source waveforms.
Recently, probabilistic methods have gained popularity.
In the statistical statespace model framework, the EEG
source localization problem is formulated as the estimation
of the posterior probability density function (pdf) of the
state based on the available observations. For the linear and
Gaussian estimation problem, the Kalman filter propagates and
updates the mean and covariance of the distribution. For nonlinear
problems and nonGaussian noise, there is no general
analytical solution to the posterior density estimation problem.
Therefore, a numerical approach is needed to evaluate the
posterior pdf of the state vector. The Particle Filter (PF) has emerged,
within the object tracking community, as one
of the most successful methods for state estimation in highly
nonlinear or nonGaussian statespace models.
PF approach will be presented as our main contribution in the ongoing research on localization of correlated EEG sources.