On-Chip Mental Stress Detection: Integrating a Wearable Behind-The-Ear EEG Device With Embedded Tiny Neural Network

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Mental stress detection is crucial for well-being and requires efficient, non-invasive methods.
• Traditional methods often rely on complex equipment or subjective reporting.
• Electroencephalography (EEG) offers high temporal resolution for brain activity analysis.

Purpose of the Study:

• To develop an efficient, wearable system for real-time mental stress detection.
• To integrate on-chip neural networks with behind-the-ear (BTE) EEG analysis.
• To identify key EEG frequency bands indicative of mental stress.

Main Methods:

• A custom wearable device captured single-channel BTE EEG signals.
• On-chip signal-to-spectrogram conversion and a compact Convolutional Neural Network (CNN) were employed.
• EEG data were collected from 15 participants during stress-inducing cognitive tasks.
• Leave-one-out cross-validation (LOOCV) and 10-fold cross-validation (CV) were used for performance evaluation.

Main Results:

• LOOCV achieved 91.72% accuracy, with an F1-score of 0.8955.
• 10-fold CV demonstrated superior performance with 95.32% accuracy and an F1-score of 0.9421 on untrained data.
• The Beta frequency band (13-30 Hz) was identified as the most significant indicator of mental stress.

Conclusions:

• The integrated BTE EEG and on-chip CNN system offers a highly accurate and efficient solution for mental stress detection.
• On-chip processing is vital for noise filtering, signal conversion, and real-time classification.
• This technology holds promise for developing advanced medical assistance tools for mental health monitoring.