Signal identification system for developing rehabilitative device using deep learning algorithms.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Human-Computer Interaction

Background:

• Increasing prevalence of neurodegenerative diseases leads to paralysis and immobility.
• Assistive technologies are crucial for individuals with impaired motor function.
• Electrooculogram (EOG) based Human Computer Interaction (HCI) offers a promising solution for communication and control.

Purpose of the Study:

• To investigate the feasibility of a nine-state HCI system using EOG signals.
• To evaluate the performance of different feature extraction techniques for eye movement classification.
• To determine the optimal model for developing an EOG-based HCI system.

Main Methods:

• Captured EOG signals using five electrodes placed around the eyes.
• Amplified and filtered signals using ADT26 bio amplifier and notch filter.
• Extracted features using reference power and band power techniques, classified with Time Delay Neural Networks (TDNN).

Main Results:

• Achieved maximum average classification accuracy of 91.09% for reference power features and 91.55% for band power features.
• Band power features with TDNN demonstrated superior performance compared to reference features across all subjects.
• Offline analysis confirmed the effectiveness of band power features with TDNN for classifying eleven distinct eye movements.

Conclusions:

• Band power features combined with TDNN models are highly suitable for classifying diverse eye movements.
• The proposed EOG-based HCI system shows significant potential as an assistive technology for paralyzed individuals.
• This approach offers a robust and accurate method for developing offline EOG-based HCI systems.