Portable deep-learning decoder for motor imaginary EEG signals based on a novel compact convolutional neural network...

Area of Science:

• Neuroscience and Biomedical Engineering
• Artificial Intelligence and Machine Learning

Background:

• Deep-learning decoders for motor imagery (MI) electroencephalography (EEG) signals typically require substantial computational resources, limiting their use in portable brain-computer interface (BCI) devices.
• The application of deep learning for MI EEG decoding in independent, portable BCI systems remains underexplored.

Purpose of the Study:

• To develop and evaluate a high-accuracy, portable deep-learning decoder for MI EEG signals.
• To demonstrate the feasibility of deploying advanced deep learning models on a single-chip microcontroller unit (MCU) for real-time BCI applications.

Main Methods:

• A novel convolutional neural network (CNN) incorporating a spatial-attention mechanism was designed for MI EEG decoding.
• The CNN model was trained on the GigaDB MI dataset and subsequently deployed on an MCU for portable operation.
• Performance was benchmarked against the EEG-Inception model, also deployed on an MCU, using the same dataset.

Main Results:

• The proposed compact CNN achieved a mean accuracy of 96.75% (±2.41%) in decoding imaginary left-/right-hand motions using 8 EEG channels.
• The EEG-Inception model achieved a lower mean accuracy of 76.96% (±19.08%) using 6 EEG channels.
• This represents the first known portable deep-learning decoder for MI EEG signals.

Conclusions:

• The developed portable system demonstrates high-accuracy deep-learning decoding of MI EEG signals, overcoming computational limitations of traditional approaches.
• This breakthrough has significant implications for the development of intelligent, wearable BCI devices, particularly for individuals with hand disabilities.
• The system's low computational cost and convenience pave the way for real-world BCI applications.