Near-lossless multichannel EEG compression based on matrix and tensor decompositions.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Data Compression

Background:

• Multichannel electroencephalogram (MC-EEG) data presents significant storage and transmission challenges due to its high dimensionality.
• Existing compression methods often struggle to efficiently capture both temporal and spatial redundancies inherent in MC-EEG signals.

Purpose of the Study:

• To develop a novel near-lossless compression algorithm for MC-EEG signals.
• To leverage matrix/tensor decomposition models for enhanced decorrelation and compression of MC-EEG data.
• To evaluate the algorithm's performance against established compression techniques.

Main Methods:

• MC-EEG data was represented in multiway (multidimensional) forms to exploit spatio-temporal correlations.
• Matrix/tensor decomposition models were analyzed for efficient decorrelation.
• A "lossy plus residual coding" approach was implemented, combining decomposition-based coding with arithmetic coding.
• The algorithm was tested on diverse scalp and intracranial EEG datasets.

Main Results:

• The proposed algorithm achieved attractive compression ratios, outperforming the separate compression of individual channels.
• For comparable compression ratios, the algorithm demonstrated a nearly fivefold reduction in average error compared to a wavelet-based volumetric MC-EEG compression method.
• The method guarantees a specifiable maximum absolute error between original and reconstructed signals.

Conclusions:

• Matrix/tensor decomposition provides an effective framework for near-lossless MC-EEG compression.
• The proposed algorithm offers a significant improvement in compression efficiency and signal fidelity.
• This approach holds promise for efficient storage and transmission of large-scale MC-EEG datasets.