Deep learning for electroencephalogram (EEG) classification tasks: a review.

Area of Science:

• Neuroscience
• Neural Engineering
• Machine Learning

Background:

• Electroencephalography (EEG) analysis is crucial in neuroscience and neural engineering, including brain-computer interfaces (BCIs).
• Machine learning and deep learning advancements are increasingly applied to EEG data for enhanced neural classification and neuroimaging.
• Robust automatic EEG signal classification is vital for practical applications and reducing reliance on expert interpretation.

Purpose of the Study:

• To systematically review deep learning applications in EEG classification.
• To identify common EEG classification tasks addressed by deep learning.
• To analyze input formulations and network architectures for deep learning models in EEG analysis.

Main Methods:

• A systematic literature review was conducted using Web of Science and PubMed databases.
• Ninety studies on deep learning for EEG classification were identified and analyzed.
• Analysis focused on task types, EEG preprocessing, input data, and deep learning architectures.

Main Results:

• Convolutional neural networks, recurrent neural networks, and deep belief networks generally outperform other architectures like auto-encoders and multi-layer perceptrons in EEG classification accuracy.
• Key application areas include emotion recognition, motor imagery, mental workload assessment, seizure detection, event-related potential detection, and sleep scoring.
• Specific input formulations, network characteristics, and classifier recommendations are detailed for each task type.

Conclusions:

• This review synthesizes current practices and performance in deep learning for EEG classification.
• Practical guidance on hyperparameter selection is provided to facilitate future research and deployment of deep learning on EEG datasets.