Understanding Mismatch Negativity with Blind Source Separation of Event-Related Potentials and Their Temporal Derivatives

Abstract

Mismatch negativity (MMN) is a difference waveform between deviant and standard event-related potentials (ERPs) extracted from scalp electroencephalography (EEG) recorded during auditory oddball stimulation. Numerous clinical studies have established MMN as a putative biomarker for several neurological conditions. However, lack of specificity and ambiguous origins cast doubt regarding whether MMN is potentially useful as a clinical instrument. In this study, we explore the origins of MMN, specifically its association with standard or deviant ERP modulation, and their temporal derivatives (dERP/dt). An intensity oddball paradigm with 80 dB standards and 70 dB and 90 dB deviants was presented to a group of 54 mantra meditators while recording their EEG from 30 scalp electrodes. Continuous EEG was processed with filtering, artifact correction, artifact rejection, epoching, and averaging to derive standard and deviant ERPs from all of the participants. Recurrent neural network (RNN) blind source separation (BSS) was employed to deconstruct ERP and dERP/dt waveforms into source signals and corresponding scalp maps. Three main sources were identified for ERPs: S1 associated with N1 and N2 peaks, S2 associated with P50 and P2 (only standards) peaks, and S3 associated with P3 (only deviants, and greater amplitude for louder deviants). Two main sources were extracted from dERP/dt data: S1 associated with negative-going frontal-central slopes, and S2 associated with positive-going frontal-central slopes. The results highlight standard-specific positivity and stimulus category specific gradients, both confirmed by source signal analysis, to be potent sources impacting MMN.