Harmonic enhancement to optimize EOG based ocular activity decoding: A hybrid approach with harmonic source...

Area of Science:

• Biomedical Engineering
• Neuroscience
• Robotics

Background:

• Intelligent robotic systems are crucial for patients with motor impairments.
• Current human-machine interface (HMI) methods often rely on limited eye-blink control, restricting system complexity.
• Decoding static ocular activities is challenging due to non-oscillatory noise like tremors and fatigue.

Purpose of the Study:

• To develop an advanced preprocessing methodology for decoding static ocular activities.
• To enhance the performance of machine learning models for electrooculography (EOG)-based HMIs.
• To explore the utility of harmonic characteristics in static ocular activities for improved classification.

Main Methods:

• A hybrid preprocessing technique combining harmonic source separation and ensemble empirical mode decomposition was implemented.
• The method targets the removal of percussive and non-oscillatory noise from static ocular movements in the time-frequency domain.
• A machine learning model utilized dual inputs: time-frequency images and vectorized features from consecutive time windows.

Main Results:

• The proposed hybrid preprocessing method achieved a 3.8% performance increase in leave-one-session-out cross-validation (LOSO) compared to baseline.
• Harmonic enhancement of high-frequency components contributed to the performance improvement.
• A strong correlation was observed between harmonic ratios in the Hilbert-Huang spectrum and LOSO performance.

Conclusions:

• The hybrid preprocessing approach effectively decodes static ocular activities for advanced HMIs.
• Leveraging harmonic characteristics of ocular signals offers a promising avenue for activity enrichment in EOG-based systems.
• This method enhances classification accuracy with minimal performance loss, paving the way for more sophisticated robotic control.