Predição de crise epiléptica pelo uso de técnicas de aprendizado de máquinas em sinais de eletroencefalograma

Abstract

Epilepsy is a brain disorder characterized by recurrent unprovoked seizures. The unpredictability of seizures negatively affects the lives of patients, causing insecurity in daily activities and may cause injury or even death. Seizure prediction can prevent, through medication or safe preparation, a series of psychological, social and physical problems that are direct consequences of this disease, such as accidents and mental disorders. This work, a proposal for online and generalized seizure prediction is presented using the microstate analysis approach and filter bank by the Discrete Wavelet Packet Transform in Electroencephalogram (EEG) signals. Both methods explored were analyzed with a reduction in the number of EEG channels. In experiments performed with only eight electrodes on the scalp, the best results achieved with eleven patients from the CHB-MIT database were a sensitivity of 100% and F P R of 0.00 h−1 , making it possible to predict an epileptic event with an average of 1,95 ± 0,76 hours in advance. This proposal contributes significantly to the development of portable device, with the reduction in the number of electrodes, capable of predicting when an epileptic seizure will occur, thus increasing the quality of life of patients with this mental disorder.