Evolutionary Games, Complex Networks and Nonlinear Analysis for Epileptic Seizures Forecasting

TL;DR

This paper introduces a novel approach combining game theory, nonlinear analysis, and machine learning to improve epileptic seizure forecasting using EEG data, with promising accuracy and real-time wearable implementation.

Contribution

It proposes a new integrated method for seizure prediction that combines dynamical network models and nonlinear time series analysis within a supervised learning framework.

Findings

Achieved 70.9% sensitivity in seizure prediction.

Increased the chance metric improvement from 42% to 50.5%.

Developing a wearable device prototype for real-time implementation.

Abstract

Epileptic seizures detection and forecasting is nowadays widely recognized as a problem of great significance and social resonance, and still remains an open, grand challenge. Furthermore, the development of mobile warning systems and wearable, non invasive, advisory devices are increasingly and strongly requested, from the patient community and their families and also from institutional stakeholders. According to the many recent studies, exploiting machine learning capabilities upon intracranial EEG (iEEG), in this work we investigate a combination of novel game theory dynamical model on networks for brain electrical activity and nonlinear time series analysis based on recurrences quantification. These two methods are then melted together within a supervised learning scheme and finally, prediction performances are assessed using EEG scalp datasets, specifically recorded for this study. Our study achieved mean sensitivity of 70.9% and a mean time in warning of 20.3%, thus showing an increase of the improvement over chance metric from 42%, reported in the most recent study, to 50.5%. Moreover, the real time implementation of the proposed approach is currently under development on a prototype of a wearable device.