Predicting Epileptic Seizures from Electroencephalography

Abstract

Epilepsy is a neurological disorder that is characterised by repeated seizures. The sudden onset of a seizure affects a patient's quality of life. Therefore, predicting an epileptic seizure in advance can improve their life by giving them warning and thus avoiding serious accidents. In this work, two general prediction models are formulated using the electroencephalography (EEG) signals of patients with Temporal Lobe Epilepsy (TLE) and Absence seizures.

EEG is the most common technique to map brain functions. Studying brain functions and how the brain regions interact is essential to understand the basis of several neurodegenerative diseases. Functional brain connectivity, as derived from multichannel EEG, is currently used as a tool to understand how the various brain regions interact with each other during a cognitive task. Researchers started to study the functional brain network by analysing the EEG data captured. Because of the high level of synchronization observed during a seizure, synchronization measures are logically the best way to assess the dynamic change in functional brain connectivity.

In the current work, Phase Locking Value (PLV), Phase Lag Index (PLI) and Synchronization likelihood (SL) were used to create functional brain connectivity networks. The networks were characterized by nine graph-theoretic parameters (assortativity coefficient, transitivity, clustering coefficient, strength of node, modularity, betweenness centrality, characteristic path length, global efficiency and radius). A machine-learning framework was used to extract the patterns that the patients' data had in common to build the prediction models.

Both general prediction models were formulated using PLI and SL connectivity networks. They achieved sensitivity (both 100%) and a false prediction rate of 0.00001/h and 0.01/h, with a maximum prediction time of 19 and 40 minutes, respectively.