A Rule-Based Classifier to Detect Seizures in EEG Signals

In this study, we develop a rule-based method for the detection of seizures in electroencephalogram (EEG) signals. The proposed method is based on the observations that EEG seizure can be modeled either as a train of impulses or as the summation of harmonically related frequency-modulated chirps. For detecting spike-train-type seizures, the proposed method estimates group delay-based features, whereas to detect seizures modeled as the summation of harmonically related frequency-modulated chirps, the instantaneous frequency-related features are extracted. Experimental results indicate that the proposed method achieves better performance than the machine learning approach in terms of total accuracy and sensitivity.