Eog and Ecg Minimization Based on Regression Analysis

A method based on regression analysis is presented that can be used for automatic minimization of EOG and ECG artifacts in the sleep EEG. INTRODUCTION It is known that the non-cortical activity gives a contribution to the EEG recordings. Electrooculogram (EOG), the electrocardiagram (ECG) and muscle activity (EMG) are the most important non-cortical sources. For sleep analysis it is important to know the undisturbed EEG. Therefore, it is necessary to detected and/or reduce non-cortical sources on EEG. Additionally, the detection/removal methods must be fully automatic. In the SIESTA project EOG, ECG as well as EEG are recorded. Therefore, it is self-evident to use these signals to reduce the influence of the EOG and the ECG. The EMG channel is not used, because the activity of different muscles is concerned. Therefore EOG and ECG signals are used for regression analysis, with the aim to reduce, at least, the linear and time-invariant part of the superposition. Elbert et al. (1985) showed theoretically how the EOG influences the scalp potentials. Berg and Scherg (1994) assumed that the EOG source is a dipole. They applied a multiple source eye correction (MSEC) method. Based on that approach Ille et al. (1997) calculated a Spatial Filter matrix with Principle Component Analysis (PCA). Lagerlund et al. (1997) applied PCA for identifying different kinds of artifacts. The artifact components were identified by visual inspection of the principle components.