Stress Classification Using Brain Signals Based on LSTM Network.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Early diagnosis of stress is crucial for preventing mental health disorders like depression.
• Electroencephalography (EEG) is a cost-effective and noninvasive method for stress detection research.

Purpose of the Study:

• To develop and evaluate a stress classification system using EEG signals.
• To compare the performance of Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM) models for stress classification.

Main Methods:

• EEG signals were recorded from 35 volunteers using a 4-electrode Muse EEG headband.
• Stress was induced using emotionally evocative movie clips, contrasted with non-stress-inducing comedy clips.
• Recorded EEG data was used to train and test MLP and LSTM classification models.

Main Results:

• A two-layer Long Short-Term Memory (LSTM) architecture was implemented for stress classification.
• The LSTM model achieved a maximum classification accuracy of 93.17% in distinguishing stress from non-stress states.
• Comparison indicated LSTM outperformed MLP in this stress classification task.

Conclusions:

• The proposed system effectively classifies stress using EEG signals.
• LSTM networks demonstrate high efficacy for automated stress detection from neurophysiological data.
• This approach holds promise for developing objective tools for mental health monitoring.