Delving α-stable Distribution For Seizure Detection And Its Application On Online Prediction

Epilepsy is a serious brain disorder, with epileptic seizures causing convulsions, loss of consciousness, and muscle spasms. Approximately 70% of diagnosed patients can have their seizures controlled effectively by antiepileptic drugs (AEDs). Surgery can provide seizure freedom for half of the remaining patients with seizures resistant to AEDs. Exploring new ways to control seizures is imperative.Currently, vagus nerve stimulation and deep brain stimulation have been applied to prevent seizures by sending regular, mild pulses of electrical energy to brain. Some mild to moderate side effects of these therapy have been reported, including hoarseness, sore throat and a cough when the device is being used (this normally occurs every five minutes and lasts for 30 seconds). Due to the side effects of electrical stimulation, responsive electrical stimulation also called closed-loop electrical stimulation is becoming one of the research hotspots in the treatment of epilepsy. This thesis focuses on responsive electrical stimulation in the treatment of epilepsy, delving robust seizure detection algorithm and its application on online prediction. The main contributions of this thesis are as follows:1) We propose a state-space model using α-stable distribution to characterize the impulse and outlying noise in EEG signals. There is serious noise in EEG caused by eye blink and muscle activities. The noise exhibits similar morphologies to epileptic seizure signals, leading to relatively high false alarms in most existing seizure detection methods. Based on a state-space model containing a nonlinear observation function and multiple features as the observations, this thesis deeply delves the effect of the α-stable distribution in the noise suppression for seizure detection from scalp EEG. Compared with the Gaussian distribution, the α-stable distribution is asymmetric and has relatively heavy tails. These properties make it powerful in modeling impulsive noise in EEG, which usually can not be handled by the Gaussian distribution. We give a detailed analysis in the state estimation process to show the reason why the α-stable distribution can suppress the impulsive noise.2) To carry out experimental study, we developed an online seizure detection closed-loop system including animal models, hardware and software platform. The developing of software platform is our work. Real-time signal can be displayed and recorded in the software platform. We can set various parameters for different requirement. It is able to detect epileptic seizure and to send pulses of electrical energy to brain for suppressing seizures.3) We give a detailed analysis to show the reason why our method can suppress the noise. To justify each component in our model, we compare our method with 4 different models with different settings and the existing approaches on a 331-hour clinical epileptic EEG data. The results demonstrate that our model is most effective in both the detection rate and the false alarm. The closed-loop system has been successfully and stably applied in the experiment for studying the suppression effect of responsive electrical stimulation.