A Deep Learning Model for Correlation Analysis between Electroencephalography Signal and Speech Stimuli.

Area of Science:

• Neuroscience
• Signal Processing
• Machine Learning

Background:

• Electroencephalography (EEG) is a non-invasive tool for brain monitoring.
• Artifacts in EEG recordings hinder accurate stimulus-response analysis.
• Current methods often use linear analysis like Canonical Correlation Analysis (CCA) to mitigate artifacts.

Purpose of the Study:

• To introduce a novel correlation analysis (CA) framework using a neural network for EEG and speech stimuli.
• To enhance the accuracy of extracting functional brain responses from EEG signals.
• To improve upon existing deep learning CA (DCCA) methods.

Main Methods:

• Developed a novel CA framework utilizing a single-layer Multilayer Perceptron (MLP) neural network.
• Designed a specific loss function to maximize the correlation between EEG signals and speech stimuli.
• Compared the proposed method against linear CCA (LCCA) and DCCA using EEG data from subjects listening to speech.

Main Results:

• The proposed neural network framework demonstrated improved correlation analysis.
• Achieved a 10.56% increase in overall Pearson correlation compared to the state-of-the-art DCCA method.
• The single MLP network approach proved effective in enhancing stimulus-response correlation.

Conclusions:

• The novel neural network-based CA framework offers a significant improvement for EEG-speech stimulus analysis.
• This approach effectively mitigates artifacts and enhances the detection of brain responses.
• The method presents a promising advancement over existing linear and deep learning techniques.