Seizure prediction and control algorithms for epilepsy prostheses

Pooja Rajdev, Purdue University

Abstract

Epilepsy is a neurological disorder triggered by excessive neuronal excitability in the brain and characterized by recurrent seizures. The sudden and apparently unpredictable nature of seizures is one of the most debilitating aspects of this disease. Methods capable of predicting the occurrence of seizures and suppressing them, would open new therapeutic possibilities, including a closed-loop epilepsy prosthesis that would abort impending seizures before their clinical onset. To that end, we exploit the electroencephalograph (EEG), and local field potentials (LFPs) for improved understanding of epileptic networks, seizure generation and their spatial and temporal characteristics. In this work, we discuss three aspects of developing an implantable device capable of controlling epilepsy: (i) epileptic seizure prediction algorithms, (ii) therapeutic and rehabilitative effects of responsive stimulation, and (iii) development of an implantable stimulator application specific integrated circuit (ASIC). An algorithm of epileptic seizure prediction is at the core of any implantable device aimed to treat the symptoms of this disorder. Linear models have been long propagated and justified for the EEG signals. Inspired by this, we develop a training-free real-time seizure prediction algorithm utilizing statistical time-series analysis. The stationarity of LFP signals is analyzed, and in this light, the method of inverse autoregressive (AR) filtering for detecting transient events is modified to adaptive inverse AR filtering. Using residue techniques, the spectral density may also be partitioned into spectral components, facilitating analysis of the alpha, beta and gamma brain rhythms. The detection of the adaptive AR algorithm is evaluated with the method of Receiver operator characteristic (ROC) curves, predictive power, and time under false positives. Furthermore, we investigate the therapeutic and rehabilitative effects of responsive stimulation on seizures induced in an animal model of epilepsy. While most patients have a single type of electrographic seizure, many others have multiple seizure patterns. It is only reasonable to assume that particular seizure patterns would reflect the type and derangement of disturbed neuronal populations along with the spatial distribution and the anatomic location of seizure focus. In this study, we determine the effects of individual stimulation parameters and test the hypothesis that additional information obtained from the seizure morphology could help optimize seizure-mitigating stimulation paradigms, maximizing the ability of the stimulation to abort a seizure. Finally, these algorithms are implemented on a TI digital signal processor and a Cambridge consultants XAP processor, enabling real-time analysis and miniaturization into hand-held or implantable systems. In order to develop less invasive and more durable deep brain stimulators, a wireless implantable ASIC is also designed and fabricated.

Degree

Ph.D.

Advisors

Irazoqui, Purdue University.

Subject Area

Biomedical engineering

COinS