Near-lossless EEG signal compression using a convolutional autoencoder: Case study for 256-channel binocular rivalry...

Area of Science:

• Neuroscience
• Computer Science
• Signal Processing

Background:

• Electroencephalography (EEG) generates large datasets due to high sampling rates and multiple electrodes.
• Storing and transmitting extensive EEG data presents significant challenges.
• Specialized compression techniques are required for efficient EEG data management.

Purpose of the Study:

• To develop a novel compression method for EEG data.
• To achieve substantial data reduction while maintaining signal fidelity.
• To create a flexible near-lossless compression scheme tailored for EEG.

Main Methods:

• Utilized a convolutional autoencoder, a type of artificial neural network, for lossy compression.
• Implemented lossless corrections based on a user-defined amplitude loss threshold.
• Applied the method to a 256-channel binocular rivalry EEG dataset for validation.

Main Results:

• The proposed method demonstrated substantial compression ratios.
• Significant improvements in compression speed were observed compared to baseline methods.
• The compression scheme proved to be flexible and near-lossless.

Conclusions:

• The artificial neural network-based compression method is effective for large EEG datasets.
• The technique offers a promising solution for efficient storage and transmission of brain activity data.
• Further research into this compression approach is warranted.