Modified-MaMeMi filter bank for efficient extraction of brainwaves from electroencephalograms

Abstract Electroencephalography (EEG) is an important tool for characterizing the functioning of the brain. Studies based on EEG involve the extraction of different spectra from EEG signals. Traditional methods of extracting these brainwaves (commonly δ, θ, α, β, γ) from EEG signals, like impulse-response filtering or wavelet decomposition, are computationally inefficient or unsuitable for real-time implementation. The Maximum-Mean-Minimum (MaMeMi) filter is a signal processing algorithm that is computationally efficient for signal filtering. The response of the MaMeMi filter is dependent on pre-decided filter coefficients. An obstacle to its implementation is that the filter coefficients have to be tuned to the sampling frequency. We propose the Modified-MaMeMi (MoMaMeMi) filter, in which the choice of coefficients is independent from the sampling frequency. Furthermore, we develop a band-pass MoMaMeMi filter which is duplicated in a filter bank, to decompose EEG signals into five common brainwaves. We validate the efficiency of the proposed filter bank by the increase in Signal-to-Noise Ratio (SNR). The maximum average increase in SNR is 19.68 dB. To prove utility of the filter-bank, we statistically compare the values of windowed average power extracted from the MoMaMeMi-filtered signals, between seizure and non-seizure components of the EEG data-set. A significant difference between the distributions suggests utility for classification problems. Since EEG-signal processing algorithms are highly customised and not limited to the 5 common brainwaves reported in this paper, we also develop a program to determine filter parameters for extraction of unique frequency bands in a bespoke MoMaMeMi filter.