Detection and evaluation of bursts in terms of novelty and surprise (Q2045505)

Many neurons fire trains of action potentials and a large number of action potentials occurring within a short period of time is referred to as a ``burst''. In this paper the authors extend the original definition of burst surprise by \textit{C. R. Legendy} and \textit{M. Salcman} [``Bursts and recurrences of bursts in the spike trains of spontaneously active striate cortex neurons'', J. Neurophysiol. 53, No. 4, 926--939 (1985; \url{doi:10.1152/jn.1985.53.4.926})] in two ways. One is to incorporate an arbitrary interspike interval distribution into the null hypothesis for computing Legendy's burst surprise, which is renamed ``burst novelty''. The other is to assign a significance measure to the obtained novelty, which is now called ``burst surprise''. They show that using a proper null hypothesis for given data is important for detecting bursts at a desired significance level; the use of an incorrect null hypothesis results in false negative or false positive burst detections, depending on the properties of the data. Their derivation of the significance measure enables them to examine quantitatively the false negative/positive effect of the use of an incorrect null hypothesis. They also demonstrate this effect in an application to real spike train data from macaque motor cortex. The result of burst detection shows false negative/positive effects depending on the irregularity of the spike train, which are consistent with the quantification based on artificial spike trains.