MAYDAY MESSAGES IN THE BRAIN?

Vasek Chvatal, Canada Research Chair in Combinatorial Optimization
Department of Computer Science and Software Engineering
Faculty of Engineering and Computer Science
	

Last June, news from Alberta hit the pages of New Scientist, the New
York Times, and other media outlets: In surveying dog-owning couples
who had a child with epilepsy, researchers led by Dr. Adam Kirton of
the University of Calgary were told that nine of the sixty dogs in the
study signalled impending seizures minutes to hours in advance,
typically by licking the child's face and whimpering. One Great
Pyrenees would even lie down on an eight-year-old ten minutes before
she had a partial seizure.

Epilepsy is a neurological disorder usually manifested by convulsive
attacks called seizures. A single seizure does not mean that a person
has epilepsy: people are diagnosed with epilepsy only if their
seizures are unprovoked and recurrent, which is the case for an
estimated 0.6% of the population. Direct and indirect costs of
epilepsy to society run to thousands of dollars per afflicted
individual per year, making the prospect of predicting seizures all
the more attractive.

In anticipating the children's seizures, the dogs may react to minute
changes in movement or smell.  Humans are not as adept at spotting
such subtle sensory stimuli, but they beat those canines hands-down
when it comes to engineering. As early as 1875, Richard Caton, a
Liverpool surgeon, published his findings on electrical impulses in the
brain. His work was the basis for subsequent experiments carried out
after World War I by Austrian psychiatrist Hans Berger.  In a paper
published in 1929, Berger introduced the term 'electroencephalogram'
(EEG) for the continuous recordings that reflect various oscillatory
patterns of electrical brain activity. Since hyper-synchronous brain
activity is the cause of seizures, the EEG is the primary tool in
diagnosing epilepsy. The world leader in computerized analysis of EEG
recordings is the Montreal company Stellate Systems. Founded in 1986,
it now distributes its hardware and software through seventeen
representatives on four continents.

The big bow-wow on top of the eight-year-old may have actually
provoked, rather than merely predicted, her seizure. Engineering
provides hope for a less controversial tool: results of research
initiated in the late 1950s and intensified since the 1980s suggest
that EEG recordings not only reflect present seizures but also contain
signs of approaching ones. Seizures are believed to develop gradually
over time, rather than emerging abruptly with no relation to the past.

Beginning as an improvised gathering at an American Epilepsy Society
Meeting in 2000, the International Seizure Prediction Group was formed
to "provide an informal structure for the major groups working in this
area to share data and ideas". On 24-28 April 2002, the First
International Collaborative Workshop on Seizure Prediction was held in
Bonn. Organized by Klaus Lehnertz of the University of Bonn and Brian
Litt of the University of Pennsylvania, the workshop was attended by
fifty-one researchers from sixteen centers in seven countries.  A
number of these participants presented evidence of EEG patterns that
discriminate intervals preceding seizures from intervals far removed
from seizures, but so far no algorithms for reliable prediction of
seizures have been designed.

When you have a hammer, everything looks like a nail. A hammer popular
with researchers in seizure prediction is the largest Lyapunov
exponent. This notion, introduced by the Russian mathematician in
1907, has been applied since the 1960s to measure the randomness of
apparently chaotic motions. It is an excellent tool for the analysis
of dynamical systems that arise in astronomy, statistical mechanics,
and fluid dynamics. It has been applied with success to fault
diagnosis, instrumentation, control, and automation. It remains to be
seen whether EEG recordings constitute another nail.

It was the consensus of the participants in the 2002 Bonn workshop
that having a large set of EEG data from patients with a variety of
epilepsy syndromes is imperative for further research in seizure
prediction. The need for a common database of such recordings was
proclaimed again in the Epilepsy and Engineering session of the
American Epilepsy Society Annual Meeting last December in New
Orleans. Collecting such a database, and making it publicly available, 
would be a fine thing.

A few links to web sites about seizure prediction are available at
http://www.cs.concordia.ca/~chvatal/neuro/epilepsy.html