A P300-Detection Method Based on Logistic Regression and a Convolutional Neural Network.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Electroencephalogram (EEG)-based brain-computer interface (BCI) systems are vital for healthcare, intelligent assistance, and recognition tasks.
• P300 detection is crucial for EEG-BCI systems, but existing algorithms struggle with cross-subject generalizability.
• Individual differences in EEG data significantly reduce the accuracy of current P300 classification methods when applied to new participants.

Purpose of the Study:

• To develop a novel classification model that addresses the lack of generalizability in existing EEG-based BCI algorithms.
• To improve the accuracy of P300 detection across different participants.
• To enhance the practical applicability of BCI systems by overcoming individual variability.

Main Methods:

• Proposed a hybrid logistic regression-convolutional neural network (LR-CNN) model.
• The model integrates an LR-based memory component to capture individual participant features.
• Incorporated a CNN-based generalization component to learn common features across participants, reducing bias.

Main Results:

• The LR-CNN model demonstrated the ability to learn and adapt to individual participant differences.
• Achieved a marked improvement in classification accuracy (>90%) for new participants compared to existing methods.
• Experimental results validated the model's effectiveness across three different datasets.

Conclusions:

• The proposed LR-CNN model significantly enhances cross-subject test accuracy for EEG-BCI systems.
• This improved generalizability is crucial for the development and widespread adoption of BCI technology.
• The model offers a promising solution for more robust and reliable brain-computer interfaces.