Spectral Spatio-Temporal Template Extraction from EEG Signals

Analysis of Event Related Potentials (ERPs) produced by brain activities can provide insight into the timing of underlying brain function. ERPs can be classified by their time/frequency characteristics and spatial location on the scalp. Traditionally, ERPs are manually located by temporally and spatially averaged EEG signals. This process is error prone and sensitive to a priori assumptions. Our proposed algorithm is a general neuroscience-focused data mining algorithm that performs time and frequency analysis on ERPs and automatically extracts templates corresponding to Spectral Spatio-Temporal (SST) regions exhibiting significant differences between experimental outcomes. The method uses time-aligned templates, which preserve the characteristics of the signal important to cognitive researchers. The ability of the selected signal templates to differentiate between stimulus responses has been verified using a pattern recognition procedure. SST template extraction is tested on data taken from a Go/NoGo task and shown to both find relationships consistent with published neuroscience literature as well as novel relationships.