Date of Award

Spring 5-25-2019

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Biomedical Engineering

First Advisor

Levi B. Good

Abstract

This study investigated the possibility of reducing the time required for accurate epileptic seizure detection through a retroactive analysis. Epilepsy is a neurological disorder affecting over 50 million individuals globally and is defined as a disorder which results in two seizures unprovoked by fever or medication. Diagnosis of epilepsy typically involves a monitored stay at an Epilepsy Monitoring Unit (EMU). The monitoring and diagnosis process ranges on average from $35,000 to $40,000 for a single stay, and the patient results from EMU are not instantly available to the patient. The collected electroencephalogram (EEG) must be analyzed by a trained EMU technician before the physician analyzes the data.

The retroactive seizure detection algorithm utilizes Teager-Kaiser energy (TE). TE increases as either a signal’s frequency or amplitude increases and is only dependent on three consecutive samples from the time-domain. The detection algorithm was trained and tested on 37,718 hours of data from 70 male Sprague Dawley rats with a total of 843 recorded seizures. The algorithm resulted in an average sensitivity of 98.1% and an average false positive rate (FPR) of 0.2660 per hour. Current algorithms involve a training stage and perform with a sensitivity between 80% and 98.8% and a FPR between 0.054 and 1 per hour. The study supports TE as a useful measure for seizure detection, and although this algorithm focuses on retroactive seizure detection, the quick response time of TE makes it well suited for real-time seizure detection.

COinS