Community Research and Development Information Service - CORDIS

Abstract

Electroencephalogram (EEG) is an important clinical tool for diagnosing, monitoring, and managing neurological disorders related to epilepsy. Neural networks provide intriguing possibilities for the analysis of the EEG. In this paper we propose a neural network based system to detect epileptic activity. The system comprises of three main components: feature extraction, feature quantization and classification. Key components in the proposed approach are the self organizing maps (SOMs) used to quantize feature vectors and the multilayer perceptron (MLP) network used to classify the quantized vectors. We performed tests with 3 sets of features: Fourier spectral energy features; wavelet energy features; Haralick's co-occurrence features. Over 96% of the epileptic activity was correctly identified with wavelet and Fourier features and with Haralick features the detection rate was in excess of 99%. Though roughly 95% of the normal activity was also correctly identified room for improvement still exists.

Additional information

Authors: VARSTA M, Helsinki University of Technology, Laboratory of Computational Engineering (FI);HEIKKONEN J, Helsinki University of Technology, Laboratory of Computational Engineering (FI);MILLÁN J del R, JRC Ispra (IT)
Bibliographic Reference: Paper presented: International Conference on Engineering Applications of Neural Networks, Stockholm (SE), June 16-18, 1997
Availability: Available from (1) as paper EN 40607 ORA
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top