Department o f Biomedical Engineering, Technical University o f Graz, Inffeldgasse 18, A-8010 Graz (Austria) ( A c c e p t e d for publication: J u n e 20, 1977)
Different m e t h o d s can be used for the d e t e c t i o n of spikes, sharp waves and o t h e r transient signals in the scalp EEG. Mostly t h e y are based on measuring the duration and sharpness of individual waves, using the second derivative (Ktonas and Smith 1974; Walter et al. 1973). Other m e t h o d s use the m a t c h e d filter (Salzberg et al. 1971), the inverse filter t e c h n i q u e (Lopes de Silva et al. 1975b) or are based on zero-crossing analysis (Ehrenberg and Penry 1976). R e c e n t l y , Gotm a n and Gloor (1976) have r e p o r t e d a m e t h o d for spike and sharp wave d e t e c t i o n by measuring sharpness and duration in relation to the background activity. We have developed a m e t h o d based on a combination o f inverse and m a t c h e d filtering. The inverse filter t e c h n i q u e is used to transform the stationary part of the EEG to a stochastically i n d e p e n d e n t time series which is a p p r o x i m a t e l y normally distributed, and the m a t c h e d filter t e c h n i q u e is used to find and classify non-stationary p h e n o m e n a in the EEG. This c o m b i n a t i o n o f the two m e t h o d s has the advantages of b o t h types of filter; namely, the background-EEGd e p e n d e n t d e t e c t i o n of non-stationary p h e n o m e n a with the inverse filter, and their classification with the m a t c h e d filter, using different types of weighting function.
Method The spike events can be considered as a linear superposition of (a) the background EEG generated by an autoregressive process and (b) one of the M spike templates o f different duration and form. The first step consists o f the inverse filter process introduced by Lopes da Silva et al. (1975a), transforming the EEG signal x i into a stochastic variable ni by the
* S u p p o r t e d by the ' F o n d s zur F o e r d e r u n g der wissenschaftlichen F o r s c h u n g in Oesterreich', project n u m b e r 3308.
formula: Min(MO,k--l) nk = xk --
~ Xk--jaj j=l
k = 1....N N -- length of the E E G time series x k = sample values of the E E G T h e n each spike t e m p l a t e si has to be transformed by the inverse filter process: MO