Correction of ocular artifacts in EEGs using an autoregressive model to describe the EEG; a pilot study

The basic idea in eye movement (EM) artifact corrections is that the actual recording is the summation of brain potentials (true EEG) and artifact. Often a regression analysis is performed, using simultaneous EEG and EOG data, to find the parameters describing the relationship between artifact and EOG derivations (EOGs). Our method uses a maximum likelihood parameter estimation and considers data from preceding sample moments as well, since there may be a delay in the artifact transferring over the scalp. For the error term (true EEG) an autoregressive function is used. Results from estimations on data from one volunteer indicate that a delay need not be considered and that 3 autoregressive parameters are sufficient. For F3 4 EOGs give only somewhat better results than 2 EOGs. For C3 and C4 2 EOGs are sufficient. For practical reasons for each of these 3 EEG recordings, 2 EOGs were used to perform corrections. Corrections were performed using either the parameters estimated for EMs and blinks together, or the parameters estimated for EMs only (used for EMs), or the parameter estimated for blinks only (used for blinks). For EMs the differences between these corrections are very small. For blinks the differences are much larger. Parameters estimated for one trial may be used to correct other trials, recorded within a period of about 15 min preceding or following that trial (cf., Fig. 6).