Magnitude quadrática da coerência na detecção da imaginação do movimento para aplicação em interface cérebro-máquina

Abstract

The brain-computer Machine (BCM) is a system that allows a person to control a device - as a computer keyboard, a wheelchair, or even prosthesis - using, for example, patterns of electroencephalogram (EEG). The identification of these patterns (event-related potentials - ERP), amidst the spontaneous brain electrical activity, has been the great challenge of the researchers. Whereas the ERP amplitude is usually of some microvolts, spontaneous activity can reach hundreds of microvolts. Some BCIs are based on movement imagination ERPs. Several techniques have been proposed to identify these ERPs in the time and frequency domain. The main problem of most of these techniques is the need for multiple epochs of EEG synchronized with movement imagination (MI), which increases the time needed to detect, making them impractical for use in BCM. The present work aims to develop algorithms to detect potentials EEG related to MI, with the fewest epochs possible for future BCM application. The magnitude-squared coherence technique (MSC), applied to EEG of left index finger movement imagination was able to detect responses mainly at central electrodes (C3, Cz and C4) in the delta band (0.1 to 1 Hz). For use of 40 epochs (M = 40), the detection probability (DP) was 57% for false alarm rate ? = 5%. The application of MSC version that uses signals from several electrodes (Multiple-MSC) allowed PRE detection with a DP = 70%. With M = 10, PD was approximately 18% for both techniques. The application of spatial filter that uses two-dimensional Laplacian operator (DF), at central electrodes (C3, C1, Cz, C2 and C4), provided the EEG signal-to-ratio improvement, resulting in the movement imagination PRE detection with a PD = 90% for M = 10. This procedure made possible the PRE detection with only one MI realization (M = 1), with a DP of 44.1%. Thus, MSC associated with the DF technique, seems to be a promising tool for the patterns detection of MI, making feasible its use in applications of ICMs.