Trained wavelets used to detect epileptic spikes

Although many authors developed different methods for automatic detection of epileptic spikes, so far a method for clinical routine has not yet evolved. Moreover the ambiguous definition of these waves makes the detection more difficult. This paper presents a new procedure to automatically detect and localise the interictal epileptic spikes based on their shape and duration. We used the discrete wavelet transform based on many trained children of a mother wavelet function to build a two-dimensional discrete wavelet spectrum. We suggest thereafter to use this spectrum in order to localise the spikes and to measure their duration. By filtering the spectrum one can easily achieve the automatic spike selection based on a duration criterion. To train the wavelets we used the conjugate gradient method commonly used until now in the field of artificial neural networks