Estimation of phase in EEG rhythms for real-time applications

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Estimating neural oscillation phase in electroencephalography (EEG) is crucial for understanding brain function.
• Accurate phase estimation is vital for developing real-time, closed-loop systems that deliver stimuli based on brain state.
• Challenges exist in phase estimation, particularly in the alpha-band, due to unknown stimulus timing or stimulus-induced signal corruption.

Purpose of the Study:

• To propose and evaluate methods for estimating the instantaneous phase of EEG rhythms at the precise time of stimulus presentation.
• To address challenges in real-time phase estimation when stimulus timing is unknown and in offline analysis when stimuli corrupt the signal.

Main Methods:

• Machine learning models were trained to learn the causal mapping from raw EEG recordings to phase estimates.
• A non-causal signal processing chain was used to generate ground-truth phase estimates for training.
• The learned causal mapping was then applied to predict phase in real-time, where non-causal methods are unsuitable.

Main Results:

• Machine learning methods demonstrated superior accuracy in estimating instantaneous EEG phase compared to traditional signal processing techniques.
• The proposed methods require only minimal pre-processing of the EEG signal.
• The approach effectively estimates phase even when the signal is corrupted by stimulus artifacts.

Conclusions:

• Machine learning offers a powerful and accurate approach for estimating instantaneous EEG phase, outperforming conventional methods.
• These advancements are critical for the development of sophisticated brain-computer interfaces and brain-state dependent paradigms.
• The ability to accurately estimate neural oscillation phase opens new avenues for research in cognitive and motor functions.