Extraction of mismatch negativity using a resampling-based spatial filtering method

Objective. It is currently a challenge to extract the mismatch negativity (MMN) waveform on the basis of a small number of EEG trials, which are typically unbalanced between conditions. Approach. In order to address this issue, a method combining the techniques of resampling and spatial filtering is proposed in this paper. Specifically, the first step of the method, termed 'resampling difference', randomly samples the standard and deviant sweeps, and then subtracts standard sweeps from deviant sweeps. The second step of the method employs the spatial filters designed by a signal-to-noise ratio maximizer (SIM) to extract the MMN component. The SIM algorithm can maximize the signal-to-noise ratio for event-related potentials (ERPs) to improve extraction. Simulation data were used to evaluate the influence of three parameters (i.e. trial number, repeated-SIM times and sampling times) on the performance of the proposed method. Main results. Results demonstrated that it was feasible and reliable to extract the MMN waveform using the method. Finally, an oddball paradigm with auditory stimuli of different frequencies was employed to record a few trials (50 trials of deviant sweeps and 250 trials of standard sweeps) of EEG data from 11 adult subjects. Results showed that the method could effectively extract the MMN using the EEG data of each individual subject. Significance. The extracted MMN waveform has a significantly larger peak amplitude and shorter latencies in response to the more deviant stimuli than in response to the less deviant stimuli, which agreed with the MMN properties reported in previous literature using grand-averaged EEG data of multi-subjects.