- author = "Otis Lkuwamy Smart",
- title = "Evolutionary Algorithms and Frequent Itemset Mining for Analyzing Epileptic Oscillations",
- school = "Electrical Engineering, Georgia Institute of Technology",
- year = "2007",
- address = "USA",
- month = may,
- keywords = "genetic algorithms, genetic programming, PSO",
- size = "approx 202 pages",
-
URL = "
http://smartech.gatech.edu/bitstream/1853/22709/1/Smart_Otis_L_200705_phd.pdf",
- abstract = "The objective of this research is to develop an automated methodology that maps epileptic networks in patients with epileptic seizures. Epilepsy afflicts over 60 million people worldwide with limited options for treatment, one-third of who have seizures that cannot be controlled by antiepileptic drugs (AED's). For 7percent of patients with epilepsy, surgery is an option because their seizures reliably begin in one discrete region. However, effective surgical procedures typically require removing a significant volume of brain tissue due to the lack of a reliable method for pinpointing the exact location of seizure onset and the pathways through which seizures are generated and spread. Consequently, medical or surgical treatment cannot control or remedy epilepsy in over 15 million individuals with the disorder, partly due to poorly localised or multi-focal seizure onsets. Although new technologies based upon preemptive electrical stimulation, the infusion of drugs, or focal cooling may be viable alternatives to surgical resection for patients with intractable seizures, these approaches still require a reliable means to map epileptic networks and determine the best sites for therapeutic intervention.",
-
abstract = "It has been suggested that certain epileptic high
frequency oscillations within the intracranial
electroencephalogram (IEEG) both between seizures
(interictal) and preceding a seizure (pre-ictal) may
localise regions important to seizure generation on the
IEEG. The work below aims to map epileptogenic tissue
by detecting epileptic oscillations in the IEEG and
statistically identifying regions of brain in which the
oscillations are concentrated. The methodology to
accomplish this aim incorporates two primary
components: 1) a module that detects the oscillations
using an feature via evolutionary algorithms; and 2) a
module that statistically filters the output of the
first module using frequent itemset mining (FIM), which
discovers chronic patterns across electrodes. Together,
these components provide decision-support for an expert
to map networks in the brain that generate seizures and
guide a suitable therapy for patients.
Prior attempts to accurately detect similar electrophysiological events in the IEEG have exhibited marginal performance. This may be because the methods lack robust choices for the selection, extraction, and combination of features that effectively distinguish epileptic events from non-epileptic events. This research successfully implements evolutionary algorithms (i.e., genetic programming, particle swarm optimisation) to detect epileptic oscillations in data with very poor signal-to-noise ratio (SNR) even after filtering and a low bandwidth up to 100 Hz.",
-
abstract = "The presented detector achieves median values near and
above 80percent sensitivity and 85percent specificity
over a sample of testing records consisting of 10-16
records with at least 30 epileptic oscillations per
record across six subjects. An analysis of variance
(ANOVA) reveals that the presented method outperforms
(in terms of sensitivity and specificity) a naive
method, which represents a standard to detect epileptic
oscillations using an energy feature and a threshold.
Permutation testing demonstrates that the method
incorporating features returned by an evolutionary
algorithm (EA) does not perform equal to random chance
at P-values near zero. The FIM clusters electrodes with
high concentrations of epileptic oscillations, where
most couplings overlap with areas of ictal (seizure)
onset and epileptogenic zones defined by clinically by
expert physicians.
These findings and results are significant, considering that this particular field of study in epilepsy is not established, meaning that this research represents some of the first efforts to investigate that epileptic high frequency oscillations and that no definitive benchmarks have been established. Optimistically, this work provides a benchmark for detecting and analysing epileptic oscillations. This research contributes an efficient means to create and fuse quality features, techniques to evaluate the quality of a feature, and an improved method to detect abnormal physiological events key to reliably mapping regions of dysfunctional brain that constitute epileptic networks. The ultimate goal is to use this new knowledge to understand the generation of seizures in these individuals and disrupt the mechanism. Hopefully, this research will lead to a better control of seizures and an improved quality of life for the millions of persons affected by epilepsy.",
- notes = "Copyright Otis L. Smart 2007",
