Processamento de sinais de atividade elétrica neuronal a partir de ferramentas matemáticas clássicas

Abstract

This dissertation aims to make a study and processing of two types of signs of neuronal electrical activity. The first was recorded from multielectrode arrays (MEA), in reference to spontaneous activity of neuronal groups grown in cultures. We analyzed two cultures considered inactive, meaning, cultures that after a few days in vitro, there was not a connection between neurons. It was applied two mathematical tools, autocorrelation and power spectral density, allowing an analysis of the signal in time domain and frequency, respectively, trying to verify whether the data recorded from these inactive cultures could be considered as noise from instrumentation. In fact, the results indicate that these signals have similar characteristics to white noise, which disturbs any computer analysis usually performed by researchers in computational neuroscience. Another source of signals used for this study were records of electroencephalography (EEG), collected on 20 patients from the Clinical Hospital of Uberlândia under prior consent. After a clinical examination by qualified professional, an analysis was performed using computational Fourier transform and power spectral density. Through amplitude graphics (Fourier transform) and spectral density, we have realized that the signal energy is concentrated around the frequencies more representative and, in some cases, there was presence of noise from power network. With the phase spectrum, it was possible to conclude that signals with similar frequencies have spectra of similar phase.