Utilização de métodos de decomposição empíricos no pré-processamento de dados de ressonância magnética funcional

Abstract

Functional Magnetic Resonance Imaging is a noninvasive technique used to map and explore brain networks through the changes in the oxyhemoglobin concentration that is caused by neural activity. One of the techniques to evaluate and measure these brain funcions is resting-state fMRI, which is indicated to subjects with some degree of neurological impairment since no cognitive task is necessary. The main problem of this exam is that it is more sensitive to noise during scanning - from physiological sources, for example, such as heart beating and breathing. The most common and hardest to correct is noise caused by a subject’s head movement. Given this fact, the objective of this thesis is to study and evaluate the effectiveness of implement empirical decomposition methods in the preprocessing stage of fMRI data. Empirical Mode Decomposition and Empirical Mean Curve Decomposition were the chosen algorithms because of their use in non-stationary and nonlinear signals. Thirty-three children participating in the ACERTA Project were classified in two groups: good readers (14 subjects) and poor readers (19 subjects). These data were submitted to five different preprocessing strategies: two for the usual preprocessing steps using or not the movements censoring; one for the Empirical Mode Decomposition method; and two for the Empirical Mean Curve Decomposition, being that one strategy uses changes proposed in this work in original algorithm. According to statistical analysis, the Empirical Mean Curve Decomposition, both the original and the modified, proved to be a promissing method for noise reduction in real fMRI data.