< Back to previous page

Project

A new generation of docoding algorithms for non-invasive Brain-Machine interfacing.

This thesis discusses three independent studies on Brain-Computer Interfaces (BCIs) for communication based on electroencephalography (EEG) andvisual stimulation. BCIs aim at establishing a direct communication channel between the brain and an external device and are therefore particularly interesting for patients whose motor output channels are severely impaired. The main objective of this doctoral project is to develop BCIs based on visual oddball paradigms (P300 component of the Event-Related Potential ERP) and Steady-State Visually Evoked Potentials (SSVEPs), toassess their performance on both healthy and disabled volunteers and tostudy possible improvements.
 
The first study (chapter 3) investigated the possibility to use Error-related Potential (ErrP) detection to improve the accuracy of P300-based spelling. When a user attemptsto communicate a symbol using a BCI, he/she receives as feedback the symbol being identified as target by the BCI. However BCIs are bound to make mistakes and therefore the provided feedback can either be correct orincorrect. In this chapter, we studied the differences between EEG responses to incorrect feedback (ErrP) and those to correct feedback as wellas the possibility to accurately detect those ErrPs in order to improvethe BCI performance. The shapes of the EEG responses to both types of feedback are consistent with earlier EEG work on error-related processes in a non-BCI context. We also showed the possibility to accurately classify those two types of EEG responses and suggested an automatic correction strategy based on ErrP detection resulting in an improved BCI performance. However, we also stressed several critical aspects of ErrP detection such as the need to collect a large amount of training data, the importance of minimizing false detection of responses to correct feedback and the minor benefit it yields to subjects that already achieve a high spelling accuracy.
 
In the second study (chapter 4), we investigated the applicability of two BCIs, one based on P300 and another one based on SSVEPs, for mental text spelling on a cohort of patients with Locked-In Syndrome (LIS). In order to be applicable for daily use, these BCI systems need to be not only accurate, but also easy to use by the patient. Their applicability therefore depends not only on the achieved performance but also on the users assessment of the mental workload associated with the BCI task and the overall satisfaction with the BCI system. For this reason, the comparison was done in terms of typing performance, mental workload, and user satisfaction. We observe a better usabilityof the SSVEP-based BCI compared to the P300-based one for the sessions performed by the tested population of locked-in patients with respect toall criteria considered. Our results suggest the advantage of developing alternative BCIs with respect to the traditional matrix-based P300 speller using different designs and signal modalities such as SSVEPs to build a faster, more accurate, less mentally demanding and moresatisfying BCI by testing both types of BCIs on a convenience sample of LIS patients.
 
In the third study (chapter 5), we investigated the possibility to successfully combine the P300 and SSVEP BCI paradigms in a single hybrid BCI. We hypothesized that such a system would allow to overcome the main limitations of each individual BCI paradigm, so as to be able to operate faster than purely P300-based BCIs and encode more commandsthan purely SSVEP-based BCIs. We collected and analyzed EEG data in a series of several experiments that allowed us to analyse the interactionsbetween the brain responses of the two paradigms, and explore the possibilities of such a hybrid BCI. We observed that the hybrid stimulation scheme did not impair any of the two types of brain responses and that the latter could be detected as accurately as in their non-hybrid counterparts. We, finally, showed the possibility to detect both types of brain responses simultaneously and discussed the feasibility of such a hybrid BCI and the gain it provides over pure P300- and pure SSVEP-based BCIs in terms of communication rate.
Date:15 Oct 2008 →  20 May 2014
Keywords:Brain-Machine interfacing (BMI), EEG (Electroencephalogram), Machine Learning, Feature selection, Wireless EEG, Patients
Disciplines:Neurosciences, Biological and physiological psychology, Cognitive science and intelligent systems, Developmental psychology and ageing
Project type:PhD project