Mental arithmetic task classification with convolutional neural network based on spectral-temporal features from EEG.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCIs) are crucial for restoring communication and motor functions in patients with severe motor impairments.
• Electroencephalography (EEG) is a widely used, non-invasive technique for measuring brain activity, making it suitable for BCI applications.
• Deep neural networks (DNNs) have shown promise in analyzing complex neural data, but often require extensive computational resources.

Purpose of the Study:

• To develop and evaluate a computationally efficient shallow neural network for EEG-based BCI.
• To assess the performance of the shallow network against deeper models for a mental arithmetic task.
• To determine the model's robustness in cross-subject classification for individuals with motor and visual impairments.

Main Methods:

• A shallow convolutional neural network (CNN) with two layers was designed to extract spectral-temporal features from EEG data.
• The shallow CNN model was compared against three other neural network architectures of varying depths.
• The models were tested on EEG data from a mental arithmetic task performed in an eye-closed state, simulating conditions for patients with motor and visual decline.

Main Results:

• The shallow CNN model achieved the highest classification accuracy at 90.68%, outperforming all other tested neural network models.
• The shallow CNN demonstrated superior robustness in cross-subject classification, with a standard deviation of only 3% compared to 15.6% for a conventional method.
• The model's efficiency in learning spectral-temporal features from EEG was highlighted.

Conclusions:

• A shallow CNN offers a highly accurate and efficient approach for EEG-based brain-computer interfaces.
• This model presents a viable and robust solution for individuals with motor disorders and visual impairments, enhancing BCI communication and function restoration.
• The findings suggest that shallow networks can be effective for complex neural data analysis, reducing computational demands.