Automatic decomposition of electrophysiological data into distinct nonsinusoidal oscillatory modes

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Neurophysiological signals are inherently noisy, nonsinusoidal, and transient, complicating the analysis of oscillatory features.
• Existing methods struggle with extracting accurate waveform shape and cross-frequency coupling, limiting our understanding of brain dynamics.

Purpose of the Study:

• To develop a novel, data-driven method for decomposing noisy, transient neurophysiological data into meaningful oscillatory modes.
• To improve the extraction and analysis of oscillatory features, such as waveform shape and instantaneous frequency, from complex neural signals.

Main Methods:

• Iterated masking empirical mode decomposition (itEMD), a flexible and fully data-driven technique based on empirical mode decomposition (EMD).
• itEMD enables single-cycle waveform dynamics extraction via phase-aligned instantaneous frequency.
• Extensive simulations and validation on multimodal, multispecies electrophysiological data.

Main Results:

• itEMD significantly enhances the separation of nonsinusoidal oscillatory components and robustly reproduces waveform shape across various conditions.
• The method successfully identified rat hippocampal θ waveform asymmetry and human occipital α oscillations without prior frequency assumptions.
• itEMD demonstrated substantially reduced mode mixing compared to existing EMD-based methods.

Conclusions:

• itEMD offers a robust and flexible approach for analyzing oscillatory dynamics in challenging neurophysiological datasets.
• By minimizing mode mixing and simplifying interpretation, itEMD facilitates advanced analyses of neural signals in relation to behavior and disease.
• This novel method advances the capability to study neural oscillations in noisy and complex biological data.