Comparison of DFT and lock-in amplifier features and search for optimal electrode positions in SSVEP-based BCI

Abstract

Brain-computer interfaces (BCIs) based on steady-state visual evoked potentials (SSVEPs) have been investigated increasingly in the last years. This type of brain signals resulting from repetitive flicker stimulation has the same fundamental frequency as the stimulation including higher harmonics.

This study investigated how the classification accuracy of a 4-class BCI system can be improved by localizing individual electroencephalogram (EEG) recording positions.

In the current work, a 4-class SSVEP-based BCI system was set up. Ten subjects participated and EEG was recorded from 21 channels overlying occipital areas. Features were extracted by applying Discrete Fourier transformation and a lock-in analyzer system. A simple one versus the rest classifier was applied to compare methods and localize individual electrode positions.

It was shown that the use of three SSVEP-harmonics recorded from individual channels yielded significantly higher classification accuracy compared to one harmonic and to the standard positions O1 and O2. Furthermore, the application of a simple one versus the rest classifier and the use of a lock-in analyzer system lead to a higher classification accuracy (mean ± S.D., about 74 ± 16%) in a 4-class BCI compared to the commonly used Discrete Fourier transformation (DFT, 62 ± 14%).

By applying a screening procedure, the optimal electrode positions for bipolar derivations can be detected. Furthermore, information about subject's specific ‘resonance-like’ frequency regions can be obtained by observing higher harmonics of the SSVEPs.

Introduction

Brain-computer interfaces (BCIs) are systems establishing a direct connection between the human brain and a computer (Vidal, 1973, Wolpaw et al., 2002) thus providing an additional communication channel. For people with severe palsy, e.g., amyotrophic lateral sclerosis or brain stem stroke such a BCI is potentially the only way to communicate with their environment.

A great variety of systems are available mostly differing in the requested mental strategy and in the type of brain signal used for classification. The majority of BCIs depend on the modulation of the mu-rhythm, for example during motor imagery (Millán and Mourino, 2003, Pfurtscheller and Neuper, 2001, Wolpaw et al., 2000). This mental strategy is accompanied by an amplitude decrease (or event-related desynchronization (ERD) (Pfurtscheller and Lopes da Silva, 1999)) of specific frequency components appearing at sensorimotor projection areas of the corresponding body parts (Pfurtscheller and Neuper, 1997). First attempts were performed using oscillatory components for the control of neuroprostheses in spinal cord injured (Müller-Putz et al., 2006). Other types of BCIs are based on the P300 of the visual event related potential and on slow cortical potentials shifts, respectively. These types of BCIs were successfully used for communication in ALS patients (Sellers and Donchin, 2006, Birbaumer et al., 1999).

A BCI system based on the detection of amplitude increases and a less demanding mental strategy may be more efficient. Based on the detection of VEP components Sutter (1992) demonstrated a BCI-system based on the visual system. Middendorf et al. (2000) introduced a system that evaluates the focus of the subject's gaze by resulting amplitude changes of the measured steady-state visual evoked potentials (SSVEPs). A flickering light source elicits SSVEPs at the corresponding flickering frequency, measurable over the occipital cortex, while the subject shifts the gaze to these stimuli. Applying two flickering light sources, the successful control of the roll position of a flight simulator was demonstrated (McMillan et al., 1995). Multiple flickering lights need to be introduced to enable higher dimensional discrimination. Recently, a system with 13 flickering light targets was presented by Cheng et al. (2002). Whenever a user gazes at certain flickering targets (visual fixation), the selected choice can be identified by the corresponding target frequency uncovered by frequency analysis.

The SSVEPs have the same fundamental frequency (first harmonic) as the stimulating frequency, but usually they also include higher (Regan, 1989) and/or sub-harmonics (Herrmann, 2001). Previous SSVEP-based BCIs were implemented on the basis of the first (Kelly et al., 2005a, Kelly et al., 2005b, Middendorf et al., 2000) or on the first and second harmonic detection (Cheng et al., 2002, Gao et al., 2003, Lalor et al., 2004). In a recent paper, we demonstrated a significant increase of the classification accuracy by using also the third harmonics (Müller-Putz et al., 2005).

In most SSVEP-BCI related publications the electroencephalogram (EEG) was derived from electrode positions O1 and O2 (international 10–20 electrode system). It was either derived bipolarly from these positions (Middendorf et al., 2000), monopolarly (Kelly et al., 2005b) or bipolarly anterior/posterior to them (Müller-Putz et al., 2005). Futhermore, in one work, high-density EEG was derived to study an SSVEP-based BCI (Kelly et al., 2005a).

Discrete Fourier transformation (DFT) had been used in most publications dealing with SSVEP-based BCIs (Cheng et al., 2002, Gao et al., 2003, Lalor et al., 2004, McMillan et al., 1995). Only in some cases a lock-in analyzer system (LAS) had been used (Middendorf et al., 2000, Müller-Putz et al., 2005).

The aim of this work was to localize optimal electrode positions and recording techniques by applying both, the DFT and LAS method, to the recorded data.