EEG-based classification of imagined digits using a recurrent neural network.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) offer alternative communication for individuals with speech impairments.
• Imagined speech recognition from electroencephalography (EEG) signals is a key area of BCI research.
• Deep learning models show promise in decoding complex neural signals.

Purpose of the Study:

• To classify imagined speech of numerical digits using EEG signals.
• To leverage temporal characteristics of EEG signals with deep learning.
• To evaluate the effectiveness of bidirectional recurrent neural networks for this task.

Main Methods:

• EEG signals were preprocessed using discrete wavelet transform for artifact removal and feature extraction.
• Multilayer bidirectional recurrent neural networks were employed for classification.
• The methodology was tested on the MindBigData database using MUSE and EPOC signals.

Main Results:

• The proposed method achieved a maximum multiclass classification accuracy of 96.18% on MUSE signals.
• Classification accuracy reached 71.60% on EPOC signals.
• The results demonstrate the model's ability to distinguish neural patterns of imagined digits.

Conclusions:

• The developed signal preprocessing and stacked bidirectional recurrent network model effectively utilize EEG's temporal resolution.
• This approach is suitable for classifying imagined digits, highlighting unique neural signatures.
• The findings support the potential of this BCI for enhancing communication for those with speech disorders.