Developing A Seizure Prediction Algorithm for A Non-Invasive Neuromodulator

In this study, a novel real-time seizure prediction algorithm is introduced to predict epileptic seizures. The proposed algorithm is expected to be applicable in a noninvasive neuromodulator. As a model of the epileptogenic zone, a small-world network of Huber-Braun neurons was built up. To assess the effects of noninvasive stimulation techniques, such as transcranial magnetic stimulation, this network was modified, and the magneto-motive forces and the electromagnetically induced currents were further applied on the network. Comprehensive investigations of the electroencephalograms of epilepsy patients have suggested that some chaotic mechanisms generate the seizures. Hence, chaos and bifurcation theory was applied, and the induced current was considered as the bifurcation parameter. The bifurcation diagram of the 'inter-spike' intervals of the mean voltage of the small world network was obtained. The precise time at which the bifurcation took place was subsequently considered as the time of the seizure onset. Comparisons of the bifurcation diagrams obtained from the patients’ electroencephalographs showed that the proposed network model could reasonably represent the actual neuronal networks of the epileptogenic zone. A dataset of the electroencephalographs of epilepsy patients and normal volunteers from an epilepsy center in Germany was used to validate the prediction algorithm. The simulation results show that the proposed algorithm has a significant capability to predict the precise occurrence of seizures and the achieved sensitivity, accuracy, and specificity of this approach were remarkably higher than those reported in previous studies.