SLES: A Novel CNN-based Method for Sensor Reduction in P300 Speller.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain Computer Interface (BCI) character spellers enable communication via eye-gazes.
• Current BCI systems require numerous sensors, limiting daily application.
• Existing sensor selection methods fail to optimally reduce sensor count without accuracy loss.

Purpose of the Study:

• To develop a novel sensor selection method for BCI character spellers.
• To reduce the number of sensors required for brain signal acquisition without compromising spelling accuracy.
• To enhance the practicality of BCI character spellers for everyday use.

Main Methods:

• A new sensor selection method utilizing a devised Convolutional Neural Network (CNN).
• A parametric backward elimination algorithm employing the CNN as a sensor ranking function.
• Experimental validation on three benchmark datasets comparing sensor reduction efficacy.

Main Results:

• The proposed method selects fewer sensors compared to existing methods across multiple datasets.
• Achieved significant sensor reduction, up to 44 sensors, while maintaining spelling accuracy.
• Demonstrated the effectiveness of the CNN-based approach in optimizing sensor subsets.

Conclusions:

• The novel CNN-based sensor selection method effectively reduces sensor requirements for BCI character spellers.
• This advancement promotes the integration of BCI technology into daily life by improving usability.
• The method offers a significant improvement over current sensor selection techniques in terms of sensor reduction.