Search for information-bearing components in neural data

Area of Science:

• Neuroscience
• Signal Processing
• Data Analysis

Background:

• Multivariate empirical mode decomposition (MEMD) is a powerful technique for analyzing multichannel data, including neural recordings.
• MEMD adaptively decomposes data into intrinsic mode functions (IMFs) that are aligned across channels and trials.
• Identifying information-bearing IMFs from noise is crucial but challenging with existing methods.

Purpose of the Study:

• To develop a robust statistical procedure for distinguishing information-bearing IMFs from noise in MEMD-processed multivariate data.
• To address the limitations of existing methods that rely on insufficient dyadic filter bank assumptions.

Main Methods:

• Proposed a statistical procedure built upon MEMD, incorporating added noise channels as a reference.
• Utilized Wasserstein distance to quantify the similarity between reference IMFs and data-derived IMFs.
• Validated the method through simulations and application to monkey cortical local field potentials during visual tasks.

Main Results:

• The proposed statistical procedure effectively identifies information-bearing IMFs.
• The method demonstrates robustness in distinguishing signal from noise in complex neural data.
• Demonstrated successful application to real-world neural recordings.

Conclusions:

• The novel statistical procedure offers a reliable approach for IMF identification in MEMD analysis.
• This method enhances the analysis of multichannel neural recordings by accurately isolating relevant oscillatory modes.
• Facilitates more precise interpretation of neural activity during cognitive tasks.