Target Detection Using Ternary Classification During a Rapid Serial Visual Presentation Task Using...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Rapid serial visual presentation (RSVP) is a high-speed brain-computer interface (BCI) paradigm.
• Target stimuli in RSVP elicit event-related potential (ERP) activity, enabling target detection.
• Variability in single-trial ERPs complicates discrimination between targets and near-non-targets, reducing BCI efficiency.

Purpose of the Study:

• To develop a novel ternary classification method to enhance ERP detection in RSVP.
• To improve the discrimination of targets from near-non-targets and far-non-targets.
• To validate the method's efficacy using magnetoencephalography (MEG) data.

Main Methods:

• A ternary classification approach was developed, differentiating targets, near-non-targets, and far-non-targets.
• An RSVP experiment was conducted using natural scene images, with targets being images containing pedestrians.
• Magnetoencephalography (MEG) data from 10 subjects were analyzed using Support Vector Machine (SVM) and Convolutional Neural Network (CNN) within the EEGNet architecture.

Main Results:

• High target detection scores were achieved with both SVM and EEGNet classifiers using MEG data.
• The ternary classification method significantly improved ERP detection scores in the EEGNet classifier by discriminating near-non-target samples.
• Visualizations revealed distinct underlying mechanisms of SVM and EEGNet classifiers in ERP detection.

Conclusions:

• Near-non-target samples in RSVP experiments exhibit separable ERP activity.
• Classifiers, particularly the EEGNet model, can enhance ERP detection scores by differentiating near- and far-non-targets based on their temporal delay relative to targets.