Análise de sinais de eletroencefalografia durante imagética motora por meio de técnicas de processamento tempo-frequência

Abstract

Motor imagery (MI) is the act of imagining a movement without actually executing it, and correlated activity can be found in the electroencephalogram (EEG). Event-related desynchronization (ERD) in the alpha band of the EEG is the most consistent phenomenon during MI, representing a reduction in power at frequencies from 8 to 13 Hz. MI EEG has been studied in the time, frequency and time-frequency domains. In this work, we investigated the temporal evolution of dynamic spectral parameters (DSPs) calculated from the short-time Fourier transform (STFT) and the discrete Hilbert transform (DHT). The DSPs calculated were: instantaneous amplitude (Ai) and instantaneous frequency (Fi) from the DHT, and root-mean-square (Arms) and center frequency (Fce) from the STFT. These parameters were calculated for EEG recordings during MI, with the aim of developing brain-machine interfaces based on easy-to-interpret measurements. These recordings are from 9 volunteers, taken from the 2A dataset of the fourth Berlin Brain-Computer Interface Competition. The experimental protocol has the following key events: a beep at t = 0 s; a visual signal prompting the subject to start the MI to be performed (4 classes of MI: right hand - RH, left hand - LH, feet and tongue) from t = 2 s to t = 3.25 s; sustained MI until t = 6 s; pause between trials lasting at least 1.5 s after t = 6 s. The results indicate the amplitude parameters (Ai, and Arms) start to increase at t = 2 s, reaching a peak around t = 3 s, decreasing until t = 4 s, and rising again at the end of the MI (t = 6 s). The frequency parameters (Fi and Fce) showed an inverted profile with respect to the amplitude parameters, with a negative correlation (corr(Ai, Fi) and corr(Arms, Fce)) present from t = 1 s to t = 6 s. Using the integral of the DSPs in the 3 to 4 s interval, statistically significant differences (SSD) were identified between the MIs in most leads and for all DSPs, with the differences between the MI of the feet compared to the MI of the tongue or RH being more evident. SSDs were also identified using the DSP integrals in the 4 to 5 s range for most leads, with the frequency DSPs complementing the amplitude DSPs, one identifying SSDs for cases where the other failed to detect, and vice-versa. As reported by other studies, it was easier to differentiate MIs related to different muscle groups (feet vs RH and feet vs tongue) than MIs related to similar muscle groups (RH vs LH).