SPECIAL BISC Seminar, 10 February 1997,320 Soda Hall, 4-5:00pm

Michael Lee (leem@cs.berkeley.edu)
Mon, 10 Feb 1997 09:22:45 +0100



Dr. Amir B. Geva1
Electrical and Computer Engineering Department
Ben Gurion University of the Negev
Beer Sheva, Israel
e-mail: geva@newton.bgu.ac.il

10 February 1997
320 Soda Hall


Key Words: Fuzzy Clustering, Time-Frequency Analysis, EEG,
Rat, Hyperbaric Oxygen

Dynamic state-recognition and event-prediction are fundamental
tasks in the field of biomedical signal processing. The problem
generally addresses s a set of ordered measurements and searches
for the recognition of some patterns in the observed elements which
will forecast an event or define a transition between two different
states of the biological system. The aim of such a prediction is to
relate early premonitory signal patterns to their probable
consequences. In this paper we present a new method for
forecasting an epileptic seizure from the EEG signal.

The study rests on the assumption that prior to the seizure there
exists a pre-ictal state (PIS) in the EEG, characterized by the
occurrence of single epileptic events and/or by a unique (unknown)
change in the background activity. Twenty five animals, chronically
implanted and wired for a dual-channel EEG recording, were
exposed to hyperbaric oxygen (HBO) until the appearance of an
electric seizure. EEG segments from the control (normobaric,
air-breathing), early and mid-exposure and the 4 minutes up to and
including the seizure were submitted to the fast wavelet transform.
The variances (energies) of the wavelet coefficients were then used
as inputs to classification by the unsupervised optimal fuzzy
clustering (UOFC) algorithm.

Applying the UOFC to a continuously sampled measurements in
semi-stationary conditions is useful for grouping discontinuous
patterns to form a warning cluster. The switches from one
stationary state to another, which are usually vague and not focused
on any particular time point, are naturally treated by means of the
fuzzy clustering. In such cases, an adaptive selection of the number
of clusters (the number of underlying semi-stationary processes in
the signal) overcomes the general non-stationary nature of the
biomedical signals.

The classification succeeded in detecting and defining several
discrete EEG states (backed by behavior) such as sleep, resting,
alert and active wakefulness, as well as the seizure. In 16 instances
it also succeeded in detecting a PIS and thus in forecasting the
seizure, at times ranging between 0.7 and 4 minutes prior to its
onset. We believe that by adding more EEG channels to the feature
extracting process and other biological signals to the classification -
the sensitivity, specificity and universality of the method could be
further improved. Universality may not be crucial by if use is made
of a dynamic version of the UOFC which will be taught the
individual's normal vocabulary of EEG states and then be expected
to detect unspecified new ones.

Michael A. Lee
Berkeley Initiative in Soft Computing
387 Soda Hall                                      Tel: +1-510-642-9827
Computer Science Division                          Fax: +1-510-642-5775
University of California                    Email: leem@cs.berkeley.edu
Berkeley, CA 94720-1776 USA       WWW: http://www.cs.berkeley.edu/~leem