Probabilistic neural network versus cubic least-squares minimum-distance in classifying EEG signals

  • Authors:

  • I. Kalatzis;N. Piliouras;E. Ventouras;I. Kandarakis;C. C. Papageorgiou;A. D. Rabavilas;D. Cavouras

  • Affiliations:

  • Department of Medical Instrumentation Technology, Technological Educational, Institution of Athens, Ag. Spyridonos Street, Egaleo GR-122 10, Athens, Greece;Department of Medical Instrumentation Technology, Technological Educational, Institution of Athens, Ag. Spyridonos Street, Egaleo GR-122 10, Athens, Greece;Department of Medical Instrumentation Technology, Technological Educational, Institution of Athens, Ag. Spyridonos Street, Egaleo GR-122 10, Athens, Greece;Department of Medical Instrumentation Technology, Technological Educational, Institution of Athens, Ag. Spyridonos Street, Egaleo GR-122 10, Athens, Greece;Psychophysiology Laboratory, Eginition Hospital, Department of Psychiatry, Medical School, University of Athens, Greece;Psychophysiology Laboratory, Eginition Hospital, Department of Psychiatry, Medical School, University of Athens, Greece;Department of Medical Instrumentation Technology, Technological Educational, Institution of Athens, Ag. Spyridonos Street, Egaleo GR-122 10, Athens, Greece

  • Venue:
  • ICCMSE '03 Proceedings of the international conference on Computational methods in sciences and engineering
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

The purpose of the present study is the implementation of a classification system for differentiating healthy subjects from patients with depression. Twenty-five depressive patients and an equal number of gender and aged-matched normal controls were evaluated using a computerized version of the digit span Wechsler test. Morphological waveform features were extracted from the digitized Event-Related Potential (ERP) signals, recorded from 15 scalp electrodes. The feature extraction process focused on the P600 component of the ERPs. The designed system comprised two classifiers, the probabilistic neural network (PNN) and the cubic least-squares (CLS) minimum-distance, two routines for feature reduction and feature selection, and an overall system evaluation routine, consisting of the exhaustive search and the leave-one-out methods. Highest classification accuracies achieved were 96% for the PNN and 94% for the CLS, using the 'latency/amplitude ratio' and 'peak-to-peak slope' two-feature combination. In conclusion, employing computer-based pattern recognition techniques with features not easily evaluated by the clinician, patients with depression could be distinguished from healthy subjects with high accuracy.