Quantifying mode mixing and leakage in multivariate empirical mode decomposition and application in motor...

Area of Science:

• Signal Processing
• Biomedical Engineering
• Machine Learning

Background:

• Noise-assisted multivariate empirical mode decomposition (NA-MEMD) is susceptible to mode mixing and leakage in intrinsic mode functions (IMFs) due to improper noise channel selection.
• These issues degrade performance in applications like motor imagery-based brain-computer interfaces (BCIs).
• Existing methods lack a way to quantify mode mixing and leakage without prior signal knowledge.

Purpose of the Study:

• To propose a novel measurement, the ML-index, for quantifying mode mixing and leakage in NA-MEMD IMFs.
• To validate the ML-index using synthetic signals and electroencephalography (EEG) data from motor imagery experiments.
• To optimize NA-MEMD parameters using the ML-index and improve BCI classification performance.

Main Methods:

• Developed the ML-index, a novel metric to quantify mode mixing and leakage in IMFs, requiring no prior knowledge of signal components.
• Applied NA-MEMD to synthetic multivariate signals and EEG recordings from motor imagery tasks.
• Optimized NA-MEMD noise channel parameters based on ML-index values and compared BCI classification performance against non-optimal parameters and conventional filtering.

Main Results:

• The ML-index effectively quantified mode mixing and leakage in synthetic signals, demonstrating its ability to improve component extraction accuracy.
• Optimizing NA-MEMD parameters with the ML-index significantly enhanced BCI classification performance on EEG data from motor imagery experiments.
• The study successfully provided a method for IMF mode mixing quantification and NA-MEMD parameter optimization.

Conclusions:

• The proposed ML-index is a valid and effective tool for quantifying mode mixing and leakage in NA-MEMD.
• Optimizing NA-MEMD parameters via the ML-index leads to significant improvements in BCI performance for motor imagery.
• This work offers a practical solution for enhancing the reliability and accuracy of NA-MEMD in signal processing applications.