Inter-subject transfer learning with an end-to-end deep convolutional neural network for EEG-based BCI

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Deep learning (DL) has shown promise in brain-computer interface (BCI) systems.
• Inter-subject classification in cognitive BCI using DL remains a challenge.
• Detecting attentive mental states from electroencephalography (EEG) data is crucial for advanced BCI applications.

Purpose of the Study:

• To propose a deep convolutional neural network (CNN) framework for detecting attentive mental states from single-channel raw EEG data.
• To evaluate the performance of different EEG input representations within the deep CNN.
• To implement inter-subject transfer learning for practical BCI applications and analyze learned attention patterns.

Main Methods:

• Developed an end-to-end deep CNN to decode attentional information from EEG time series.
• Investigated the impact of three different EEG representations on CNN performance.
• Employed inter-subject transfer learning for classification to avoid re-training.
• Visualized and analyzed network perceptions of attention and non-attention states.

Main Results:

• Achieved an average classification accuracy of 79.26% for attention detection.
• Demonstrated that only 15.83% of subjects fell below the 70% accuracy threshold.
• Showcased superior performance compared to conventional classification methods for attention detection.
• Confirmed that learned patterns from raw EEG data are meaningful through visualization.

Conclusions:

• The proposed framework significantly enhances attention detection accuracy, particularly with inter-subject classification.
• This study advances end-to-end learning research by enabling networks to learn from raw data with minimal pre-processing.
• The framework reduces computational load by eliminating extensive data preparation and feature extraction, making it practical for BCI development.