Learning event-related potentials (ERPs) from multichannel EEG recordings: A spatio-temporal modeling framework with a fast estimation algorithm

Extracting event-related potentials (ERPs) from multichannel EEG recordings remains a challenge due to the poor signal-to-noise ratio (SNR). This paper presents a multivariate statistical model of ERPs by exploiting the existing knowledge about their spatio-temporal properties. In particular, a computationally efficient algorithm is derived for fast model estimation. The algorithm, termed SIM, can be intuitively interpreted as maximizing the signal-to-noise ratio in the source space. Using both simulated and real EEG data, we show that the algorithm achieves excellent estimation performance and substantially outperforms a state-of-the-arts algorithm in classification accuracies in a P300 target detection task. The results demonstrate that the proposed modeling framework offers a powerful tool for exploring the spatio-temporal patterns of ERPs as well as learning spatial filters for decoding brain states.