Spike detection in EEG by LPP and SVM

This study presents a computer algorithm to detect epileptiform discharges (spikes) in electroencephalography (EEG) that are manifestations of an epileptogenetic abnormality of the brain. Visual analysis is rater-dependent and time consuming, especially for long-term recordings, such as in sleep studies or in ambulatory EEG. Computerized methods can improve efficiency in reviewing long EEG recordings. The proposed method applies coarse to detailed modeling of the spike waveform and classifies the transients based on Locality Preserving Projections (LPP) and Support Vector Machines (SVM). The method achieves high sensitivity with low false positive rate in a intra-patient cross-validated setting and thus constitutes a valuable tool for automatic spike assessment.