Abstracts

Performance Characteristics of an Automated Seizure Warning Algorithm (ASWA) Utilizing Dynamical Measures of the EEG Signal and Global Optimization Techniques.

Abstract number : 1.123
Submission category :
Year : 2001
Submission ID : 140
Source : www.aesnet.org
Presentation date : 12/1/2001 12:00:00 AM
Published date : Dec 1, 2001, 06:00 AM

Authors :
J.C. Sackellares, M.D., Neurology and Neuroscience, University of Florida and Gainesville VAMC, Gainesville, FL; L.D. Iasimidis, Ph.D., Bioengineering, Arizona State University, Tempe, AZ; P.M. Pardalos, Ph.D., Industrial and Systems Engineering, Universi

RATIONALE: Approximatey 91% of complex partial and secondarily generalized seizures are preceded by a dynamical state transition which is detectable through analysis of EEG dynamics (J. Combinatorial Optimization 2001; 5:9, American Mathematical Society 1999; vol. 55, World Scientific; Singapore; 1991). It is characterized by convergence, among critical cortical sites, of the value of STLmax, a measure of signal stability (dynamical entrainment). We have developed an ASWA which employs integer quadratic optimization, a global optimization algorithm, to identify critical electrode sites and warns of an impending seizure when these sites become dynamically entrained. We report the sensitivity, specificity, and false prediction rate of the ASWA.
METHODS: Continuous 28-channel long-term intracranial EEG recordings previously obtained in 5 patients with medically intractable partial seizures was used to test the ASWA. Patient 1 had 24 seizures in a 10-day recording; patient 2 had 19 seizures in 6 days; patient 3 had 7 seizures in 12 days; patient 4 had 8 seizures in 5 days; and patient 5 had 9 seizures in 4 days. The ASWA involved the following steps: (1) calculate STLmax for sequential 10.24 second epochs for each electrode site, (2) compare mean STLmax values for all possible electrode pairs, (3) select critical electrode pairs using integer quadratic optimization, (4) warn when entrainment occurs in critical electrode pairs. The warning was correct if a seizure occurred within 3 hours.
RESULTS: The sensitivity of the ASWA in patients 1, 2, 3, 4, and 5 was 84.2%, 87.5%, 85.7%, 83.3% and 87.5%, respectively. The specificity was 99.936%, 99.932%, 99.944%, 99.979% and 99.929% for patients 1-5 respectively. The false prediction rate was 0.08, 0.20, 0.05, 0.06, and 0.15 per hour. The average intervals between warning and subsequent were 78.8[plusminus]12.4, 66.8[plusminus]19.3, 56.5[plusminus]22.2, 96.9[plusminus]33.4, and 62.9[plusminus]25.3 minutes for patients 1-5, respectively.
CONCLUSIONS: This ASWA is can warn of an impending seizure with performance characteristics that could have practical clinical utility. The ASWA could be incorporated into clinical EEG monitoring system or, incorporated into an implantable chip and used to activate timed physiological or pharmacological interventions.
Support: Grants from the NIH, NSF, U.S. Department of Veteran[ssquote]s Affairs, and University of Florida Biomedical Engineering Program.