Learning efficient codes from epileptic EEG
Leticia Cabral Correia1; Gean Carlos Lopes de Sousa2; Carlos Magno Sousa Junior2; Marta de Oliveira Barreiros2; Fausto Lucena de Oliveira2; Allan Kardec Barros2; Noboru Ohnishi3
1 Instituto Federal do Maranhão; 2 Universidade Federal do Maranhão; 3 Nagoya University
doi:10.20906/CPS/CILAMCE2017-0698
Resumo
Epilepsy is a neurological condition characterized by recurrent seizures, caused by brief disturbances in the electrical functions of the brain. Epilepsy affects people in all age groups, regardless of gender, race or financial condition. Several studies are developed in the field of neuroscience with the object of the most diverse aspects of epilepsy, from causes to treatment. The electroencephalogram (EEG), used in the diagnosis and monitoring of the disease, is an examination that records the brain's electrical activity and can be done through electrodes implanted in the scalp or directly into the brain. The EEG of an epileptic patient is divided into four stages: interictal, crisis-free period; Ictal, period of the crisis; Preictal period immediately preceding a crisis; And postictal, immediately following a crisis. This work analisys the EEG signal based on a theoretical viewpoint from a strategy called redundancy reduction. The efficient coding hypothesis, proposed by Horace Barlow in 1961, suggests that neurons act in a way that minimizes statistical redundancies in the transmission of information. In this work, efficient coding was done through Independent Component Analysis (ICA), which is a computational method that allows the decomposition of a signal into statistically independent components. A total of 15 hours of interictal signal and 6 hours of pre-ictal signal were segmented in one-second periods and used as input to the FastICA algorithm. In the basis functions obtained for each period, the time-frequency analysis was performed. Our results show that the efficient coding could be used to explain the neural processing subserving the genesis of the epileptic seizures. The time-frequency analysis of the characteristic basis functions of the interictal period show that there are the presence of components at high frequencies (> 58 Hz) at this stage. In the basis functions of the preictal period, the components at high frequencies are less frequent and arise less regularly. This difference may indica
Palavras-chave: Epilepsy; Seizures; Efficient Coding