A histogram-based electroencephalogram spike detection

This paper aims to improve the performance of a proposed electroencephalogram (EEG) spike detection technique. This technique accentuates the signature of spikes in the time domain signal using a nonlinear energy operator by amplifying high frequency activities such as spikes. The resulted signal is convolved with a smoothing window to reduce the effect of noise. Then, values of the resulted signal higher than a threshold value are considered as spikes. The instantaneous nature of the technique and its very low computation make it an ideal tool for spike detection. In this approach selection of the threshold value is crucial for the accuracy of the technique. This paper is aimed at improving the technique using a new approach for the threshold selection using the histogram of the smoothed nonlinear energy operator. The efficiency of the presented spike detection method has been evaluated using both synthetic signals and real newborn EEG. Results in fits paper show that the proposed technique is superior to the original technique both in terms of sensitivity and selectivity.