Dual branch neural network with dynamic learning mechanism for P300-based brain-computer interfaces.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interface (BCI) systems provide alternative communication for individuals with disabilities.
• P300 spellers are reliable BCI systems but suffer from class imbalance, affecting performance.
• Current class rebalancing methods for P300 spellers have limitations in ensuring balanced output.

Purpose of the Study:

• To propose a novel dual branch learning (DBL) method to mitigate class imbalance in P300 speller BCI systems.
• To improve feature representation and classification performance by addressing inherent data imbalances.
• To enhance the reliability and accuracy of BCI systems for users with motor impairments.

Main Methods:

• Developed a dual branch learning (DBL) method incorporating uniformly sampled and reverse-sampled data.
• Implemented a dynamic learning mechanism to progressively emphasize minority class samples (P300 component).
• Evaluated the DBL method on public and self-collected datasets using a subject-dependent approach.

Main Results:

• The DBL method achieved high classification accuracies of 97.37% and 88.72% on tested datasets.
• Demonstrated superior and more reliable performance compared to existing deep learning and rebalancing techniques.
• Effectively addressed the class imbalance issue inherent in P300 BCI datasets.

Conclusions:

• The proposed DBL framework shows significant promise for enhancing P300-based BCI systems.
• DBL offers a robust solution for improving classification accuracy and reliability in BCI applications.
• This method provides a valuable advancement for assistive technologies utilizing Brain-Computer Interfaces.