Real-time noise cancellation for EEG monitoring and brain-computer interfaces

EEG recordings carry a low signal-to-noise ratio, which hampers the effectiveness of EEG monitoring, brain-computer interfaces, and brain-state-dependent stimulation. To address this, researchers have developed a real-time noise-canceling algorithm named rtSOUND. By adapting an existing offline algorithm, SOUND, to work in real-time, rtSOUND processes EEG data using a spatial filter instead of conventional temporal filtering.

This spatial approach allows for each data sample to be cleaned individually in near real-time, making it well-suited for real-time applications. Unlike methods relying on frequency content assumptions, rtSOUND assumes that noise topographies differ from brain signals. Using an EEG forward model, rtSOUND cross-validates the EEG time series and estimates the spatial distribution of noise, resulting in a linear spatial filter that cleans the EEG data.

The algorithm processes each streamed EEG sample synchronously and updates the filter asynchronously in parallel. This ensures that noise reduction is immediate and accurately adapts to the current EEG state. Validated with TMS-EEG and resting-state EEG datasets from 25 healthy subjects, rtSOUND consistently improved SNR across various metrics.

It showed particular benefits for channels with poor SNR, while preserving waveforms in high-quality channels. With processing speeds capable of handling data streamed at 5 kHz, rtSOUND provides a robust method for real-time EEG signal enhancement, improving signal quality for brain-state-monitoring applications.