| Interictal epileptic spikes can reveal important information about the generators of epileptic activity, especially during the pre-surgical investigation of intractable epilepsy. The visual inspection of EEG recordings done by neurologists to collect the spikes is time consuming and might not always be exhaustive, especially regarding spikes of small amplitude. Automatic spike detection saves a lot of the neurologist's time. Exhaustive detection of spikes is even more necessary when these events are used in an event related paradigm to detect hemodynamic activity, (functional Magnetic Resonance Imaging, Near infraRed Spectroscopy). In this context, this study proposes a new semi-automatic spike detection , approach that aims to make an inventory of all the spikes in a run accordingly to few samples marked by an experienced neurophysiologist. The method integrates information from the neurologist about time (duration), space (correlations between EEG electrodes) and frequencies (spectral characteristcs). Based on Empirical Mode Decomposition (EMD) representation of the signals, the method is first validated with realistic simulations. We then present results for three patients for which this algorithm exhibits a large number of spike like activities of small amplitude but similar sources as the samples given by the neurologist.Keywords: Biomedical, signal processing, electroencephalography, empirical mode decomposition, epilepsy, spikes detection. |
