EEG-based imagined words classification using Hilbert transform and deep networks

Decoding Brain’s Electrical Activity: Leveraging Hilbert Transforming Techniques for EEG Analysis

“…The Hilbert Transform is a fundamental tool in signal processing, particularly valued for its application in analyzing EEG signals [41]. This mathematical transform derives the analytic representation of a real-valued signal, enabling the extraction of its amplitude envelope and instantaneous phase [42]. Utilizing the Hilbert Transform in EEG signal analysis facilitates a deeper understanding of brain activity's complex, oscillatory nature, offering insights into the amplitude and phase dynamics of neural oscillations across different brain states and conditions [43].…”

EEG Amplitude Modulation Analysis across Mental Tasks: Towards Improved Active BCIs

“…The PSD was calculated from each of these windows using Welch's method, employing the MNE toolbox [56], where the power of each frequency was normalized by the sum of the entire power spectrum. PSD features were extracted from several conventional frequency bands, including theta (4-8 Hz), alpha1 (8-10 Hz), alpha2 (10-12 Hz), beta1 (12-21 Hz), beta2 (21)(22)(23)(24)(25)(26)(27)(28)(29)(30), theta to beta , and delta to gamma (0-50 Hz), for each EEG electrode. In addition to the PSD features, we also extracted power ratios, notably alpha (8-12 Hz) to beta (12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) and theta to beta ratios, for each electrode.…”

“…PSD features were extracted from several conventional frequency bands, including theta (4-8 Hz), alpha1 (8-10 Hz), alpha2 (10-12 Hz), beta1 (12-21 Hz), beta2 (21)(22)(23)(24)(25)(26)(27)(28)(29)(30), theta to beta , and delta to gamma (0-50 Hz), for each EEG electrode. In addition to the PSD features, we also extracted power ratios, notably alpha (8-12 Hz) to beta (12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30) and theta to beta ratios, for each electrode. Subsequent to the extraction process, all features underwent log-scaling.…”

“…The process involved the decomposition of EEG signals into five spectral bands utilizing a filter bank (FB). The bands selected for this study included the conventional delta (1-4 Hz), theta (4-8 Hz), alpha (8-12 Hz), beta (12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26)(27)(28)(29)(30), and gamma (30-50 Hz), given their established roles in representing neuronal dynamics [57]. We used a zero-phase FIR filter to execute the signal filtering, which consisted of two successive filtering steps in opposing directions on a mirror-padded version of the input signal.…”

“…Other approaches have included the development of new signal processing and feature extraction tools to help sift out important brain patterns from noise. For example, the last decades have seen developments in features such as fractal dimension and entropy measures [23,24], as well as the use of amplitude envelope-based features [25,26] to complement traditional spectral power [27][28][29], spectral coherence [30,31], and time domain statistics [32] features. More recently, deep learning models have emerged as data-driven methods that can help to advance BCI technologies [33][34][35][36].…”