Document Type
Thesis
Date of Award
12-31-1991
Degree Name
Master of Science in Biomedical Engineering - (M.S.)
Department
Biomedical Engineering Committee
First Advisor
Stanley S. Reisman
Second Advisor
Elizabeth Pinkhasov
Third Advisor
David S. Kristol
Abstract
EEG analysis of epileptic patients is extremely time consuming. Patients are monitored up to 36 hours at a time. The neurologist must visually review stacks of EEG records to determine whether they contain epileptic or artifactual activity. This process is very subjective and can vary from one neurologist to another based on his or her training and experience. The neurologist is only human and may tire or lose his or her concentration. Automated EEG analysis can efficiently reduce the amount of data and time involved in analysis. It is also more consistent and objective than human analysis.
The main objective of this thesis is to develop an algorithm which will detect epileptic spikes (off-line) from EEG data. The first algorithm is based on a parametric method where the EEG signal, a stationary signal, is predicted by past signals. This autoregressive filter should remove any unexpected events (non-stationaries) such as epileptic spikes. Any large error signal (EEG signal minus the predicted signal) should represent epileptic spikes which are considered to be non-stationary events.
The second algorithm is based on a mimetic approach which tries to imitate the neurologist's analysis process. Decision logic, which is based on parameter thresholds and a prior knowledge of a spike, determines whether epileptic spikes exist.
Six, two minute or less, EEG records were visually analyzed (epileptic spikes were scored) and used as input to the two algorithms. The EEG records were taken from patients who had been monitored for possible epilepsy.
The autoregressive filter was not reliable because it did not detect epileptic spikes. Many false positives and false negatives were detected by this algorithm. It seems that the EEG signals were not stationary enough for this filter to work correctly. The mimetic approach was much more successful and did detect the epileptic spikes in the EEG records.
Recommended Citation
VanderSchraaf, Walter, "Automated EEG analysis and spike detection" (1991). Theses. 2638.
https://digitalcommons.njit.edu/theses/2638
