Stimulus-aware spatial filtering for single-trial neural response and temporal response function estimation in high-density EEG with applications in auditory research

Abstract

AbstractObjective. Neural responses recorded using electroencephalography (EEG) and magnetoencephalography (MEG) can be used to study how our brain functions, as well as for various promising brain computer interface (BCI) applications. However, a common problem is the low signal to noise ratio (SNR) which makes it challenging to estimate task-related neural responses or the temporal response function (TRF) describing the linear relationship between the stimulus and the neural response, particularly over short data windows. To address these, we present an algorithm that takes advantage of the multi-channel nature of the recordings, and knowledge of the presented stimulus, to achieve a joint noise reduction and dimensionality reduction using spatial filtering. Methods. Forward modeling is used to project the stimulus onto the electrode space. The second-order statistics of this estimated desired signal and the raw neural data are used to estimate spatial filters that maximize the SNR of the neural response, based on a generalized eigenvalue decomposition. Main Results. 1. For synthesized EEG data, over a range of SNRs, our filtering resulted in significantly better TRF estimates from 20 s trials, compared to unfiltered EEG data. 2. On a dataset from 28 subjects who listened to a single-talker stimulus, our method resulted in correlations between predicted neural responses and the original EEG data that were significantly higher compared to standard approaches. 3. On a dataset of 16 subjects attending to 1 speaker in a two-speaker scenario, our method resulted in attention decoding accuracies which were higher compared to existing forward modelling methods. Significance. Our algorithm presents a data-driven way to denoise and reduce dimensionality of neural data, thus aiding further analysis, by utilizing the knowledge of the stimulus. The method is computationally efficient, and does not require repeated trials, thereby relieving experiment design from the necessity of presenting repeated stimuli to the subjects.