Accession Number:

ADA295629

Title:

Neural Networks in Seizure Diagnosis.

Descriptive Note:

Final rept.,

Corporate Author:

ARMY ARMAMENT RESEARCH DEVELOPMENT AND ENGINEERING CENTER WATERVLIET NY BENET LABS

Personal Author(s):

• Johnson, M. A.
• Kendall, G.
• Cote, P. J.
• Meisel, L. V.

Report Date:

1995-03-01

Pagination or Media Count:

17.0

Abstract:

A monitor has been designed to detect the onset of status epilepticus associated with complex partial seizures in children. A unique sensor technology was developed to detect the minor, barely perceptible tremors characteristic of partial seizures. A microcontroller analyzes the sensor data and activates a remote tetherless alarm when a seizure is detected. However, the sensor response is similar for both casual and seizure activity, therefore, false alarms do occur. Neural networks have been studied as a means of analyzing the sensor response and differentiating seizure activity from casual motion. The network uses elements of the normalized power spectrum of the response data as a feature set. Our results indicate this approach provides a faster and more reliable means of accurately detecting seizures than the method currently employed.

Descriptors:

• *NEURAL NETS
• *MONITORS
• *EPILEPSY
• DETECTORS
• POWER SPECTRA
• FALSE ALARMS
• RESPONSE
• CHILDREN
• CONVULSIVE DISORDERS
• TREMORS

Subject Categories:

• Medicine and Medical Research
• Stress Physiology

Distribution Statement:

APPROVED FOR PUBLIC RELEASE