New feature extraction approach for epileptic EEG signal detection using time-frequency distributions
Angel Navia-Vazquez and Carlos Guerrero-Mosquera
Medical & Biological Engineering & Computing
This paper describes a new method to identify
seizures in electroencephalogram (EEG) signals using
feature extraction in time frequency distributions (TFDs).
Particularly, the method extracts features from the
Smoothed Pseudo Wigner-Ville distribution using tracks
estimated from the McAulay-Quatieri sinusoidal model.
The proposed features are the length, frequency, and
energy of the principal track. We evaluate the proposed
scheme using several datasets and we compute sensitivity,
specificity, F-score, receiver operating characteristics
(ROC) curve, and percentile bootstrap confidence to conclude
that the proposed scheme generalizes well and is a
suitable approach for automatic seizure detection at a
moderate cost, also opening the possibility of formulating
new criteria to detect, classify or analyze abnormal EEGs.