Applications of Neural Mass Model-Based Cubature Kalman Filter to Chronic Seizure

Abstract

Temporal lobe epilepsy is a common neurological disease to which about 30% of patients cannot control the seizures by medications Moreover tonic-clonic seizure is the leading cause of the sudden unexpected death in epilepsy a fatal complication of epilepsy Thus there is an urgent need to develop a system that can predict or detect the onset of seizures The neuron mass model is a nonlinear model capable of interpreting the dynamics of neuron reception and sensation Not only the initiation and termination of epileptic activities but also the unforeseeable onset could be simulated by the model The model may be utilized for seizure prediction detection or even exploring the mechanisms of seizure suppression The purpose of this study is to develop a seizure detection algorithm based on estimation of the parameter of the neural mass model by a nonlinear constrained square root cubature Kalman filter The pilocarpine-treated chronic seizure mice were used to verified the proposed filter The results from 19 spontaneous recurrent seizures show that the proposed filter can indicate seizures 13 3±13 4 seconds before the beginning of tonic-clonic seizures or 3 9±11 5 seconds before the onset of seizures with a mean accuracy of 94 2 % sensitivity of 84 0 % and false alarm rate of 4 5% In conclusion the estimation of parameter and the state variables has been achieved in silico and both prediction and detection could be accomplished by the model-based cubature Kalman filter

Date of Award	2020
Original language	English
Supervisor	Ming-Shaung Ju (Supervisor)