A test of four EOG correction methods using an improved validation technique Trieu T.H. Pham a , Rodney J. Croft b, ⁎, Peter J. Cadusch c , Robert J. Barry b a Brain Sciences Institute, Swinburne University of Technology, Victoria 3122, Australia b School of Psychology, University of Wollongong, Wollongong 2522, Australia c Centre for Atom Optics and Ultrafast Spectroscopy, Swinburne University of Technology, Victoria 3122, Australia abstract article info Article history: Received 21 May 2010 Received in revised form 19 October 2010 Accepted 19 October 2010 Available online 27 October 2010 Keywords: Electroencephalogram Ocular artifact Eye movements EOG correction methods Validation technique Event related potential (ERP) Auditory N100 Peak Difference A group of methods that are employed for removing ocular artifact from the electroencephalogram (EEG) is referred to as electrooculogram (EOG) correction methods. These use least-square linear regression, and the relative success of these is yet to be established. Improving on previous EOG correction validation studies, we present a new validation technique (with greater face validity) and use it to compare four commonly employed EOG correction methods. Data consisted of ERP traces to auditory stimuli that were embedded in up, down, left and right eye movements (EMs), recorded from 24 subjects. A ‘Peak Difference’ validation measure was employed, which determined the magnitude of the difference of two auditory N100 peaks (those associated with EMs with opposing polarities). All correction methods produced data that was better than not correcting at all. EOG correction methods that accounted for vertical EM, horizontal EM and blink artifact separately using separate EOG channels, produced the best corrections, with some further advantage in methods that enhanced signal (EOG) to noise (EEG) ratios when calculating correction coefficients. © 2010 Elsevier B.V. All rights reserved. 1. Introduction Ocular artifacts in the electroencephalogram (EEG) are common and can severely distort the EEG. A conservative approach is to reject segments of EEG that are contaminated by ocular artifact. This approach often results in the loss of a considerable amount of EEG data (Small, 1971; Krishnaveni et al., 2005). Similarly, fixation instructions distort EEG data due to the relation between cognition and eye movement (Verleger, 1991), and are thus problematic for EEG analysis. An alternative approach is to use regression-based electro- oculogram (EOG) correction methods such as proposed by Verleger et al. (1982), Gratton et al. (1983), Semlitsch et al. (1986) and Croft and Barry (2000a), or data separation methods such as Brain Electrical Source Analysis (BESA; Berg and Scherg, 1991a), Principle Component Analysis (PCA; Berg and Scherg, 1991b), and Independent Component Analysis (ICA; Makeig et al., 1996). These methods can effectively (but not equally) account for a substantial amount of ocular artifact in the EEG. Certain regression-based EOG correction methods account for ocular artifact by subtracting weighted portions (correction coefficients; Bs) of ocular voltage (EOG) from the EEG (hereafter ‘EOG correction methods’). Data separation methods account for ocular artifacts by identifying the artifact ‘components’ and removing them from the EEG. While no consensus as to the optimal method has been reached, Croft et al. (2005) reported that within EOG correction methods, there were clear differences in the success of different methods. The present study aimed to improve the validation technique employed by Croft et al. (2005), providing a more easily interpretable validation technique and testing a new data set. Other comparisons, such as within the data separation methods, and between data separation and EOG correction methods, are beyond the scope of the present study. The literature contains a number of different approaches for determining whether EOG correction methods are valid. For example, Gasser et al. (1986) proposed that good correction should result in similar broadband power values for different epochs of a subject's resting EEG, and others have employed simulations to assess the issue (Croft and Barry, 1998, 2000b; van den Berg-Lenssen et al., 1994). However, the former technique is limited by the correlation that would also result from residual EOG in the EEG channels, as well as the association between ocular voltage and true neural potentials, and simulations are based on a number of assumptions that are themselves in need of determination (Croft et al., 2005). Visual inspection (Girton and Kamiya, 1973) and blind scoring (Schlögl et al., 2007) are other forms of validation that have been employed, but these too may be problematic in that small residual artifacts are difficult to identify (Whitton et al., 1978). Conversely, the validation technique of Croft et al. (2005) compared the corrected up eye movement (EM) with the corrected down EM for vertical EM (VEM) (and similarly for the left and right EMs for horizontal EM (HEM) validation). The rationale for this is that International Journal of Psychophysiology 79 (2011) 203–210 ⁎ Corresponding author. Tel.: +61 2 42213652; fax: +61 2 42214163. E-mail address: rcroft@uow.edu.au (R.J. Croft). 0167-8760/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpsycho.2010.10.008 Contents lists available at ScienceDirect International Journal of Psychophysiology journal homepage: www.elsevier.com/locate/ijpsycho