Detecção de Sinais Anormais de EEG por meio de Microestados e Aprendizado de Máquina

Abstract

The detection of abnormal electroencephalogram (EEG) signals is the first step to aid in the identification of neuropathologies, having the potential to considerably reduce the time between signal capture and medical report. A technique that has not yet been explored for this specific task, but has shown good capacity in the detection of mental disorders, due to its ability to capture spatial and temporal information, is the EEG microstate analysis. In this work, a methodology for detecting abnormal EEG signals is proposed, combining the use of microstate analysis and a Learning Vector Quantization (LVQ) network with the intention of improving the prototypes of the microstates obtained initially by the commonly used modified k-means clustering method. Experimental results in a public dataset suggest that microstate analysis, which uses the topographic characteristics of the EEG signal, is promising for the detection of abnormal EEG signals, regardless of an a priori specified neuropathology. Furthermore, the use of microstates with LVQ proved to be statistically better than the traditional method of obtaining microstate prototypes.