IENet: a robust convolutional neural network for EEG based brain-computer interfaces.

Area of Science:

• Neuroscience
• Computer Science
• Signal Processing

Background:

• Electroencephalogram (EEG) based brain-computer interfaces (BCIs) require robust signal processing for complex applications.
• Deep learning excels at feature extraction and dependency analysis in various domains.
• Existing EEG algorithms face challenges in generalizability across diverse BCI paradigms.

Purpose of the Study:

• To design a robust deep learning algorithm for analyzing EEG signals across multiple BCI paradigms.
• To introduce InceptionEEG-Net (IENet), a novel neural network ensemble for enhanced EEG analysis.
• To improve the generalizability and reliability of EEG signal processing in BCIs.

Main Methods:

• Developed InceptionEEG-Net (IENet), inspired by InceptionV4 and InceptionTime architectures.
• Utilized multi-scale convolutional layers and length-1 convolutions for rich feature extraction with fewer parameters.
• Proposed average receptive field (RF) gain to optimize detection of long patterns in EEG data.
• Evaluated IENet against state-of-the-art methods across five distinct EEG-BCI paradigms.

Main Results:

• IENet achieved generalizability comparable to state-of-the-art paradigm-agnostic models on test datasets.
• Feature explainability analysis confirmed IENet's ability to extract neurophysiologically interpretable features.
• The proposed average RF gain effectively increased the receptive field size for deep convolutional neural networks (CNNs).

Conclusions:

• InceptionEEG-Net (IENet) demonstrates robust generalization across various BCI paradigms.
• The model's capacity for extracting interpretable features enhances the reliability of BCI algorithms.
• Increasing RF size using average RF gain is crucial for deep CNNs in EEG analysis.