Exploring Deep Learning in Electrodermal Activity Recording for Continuous Detection of Pain Intensity Level.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Objective pain assessment is crucial for effective opioid treatment and addiction prevention.
• Electrodermal activity (EDA) shows potential for pain detection but is limited by intermittent measurements.
• Current methods lack continuous, objective pain intensity monitoring.

Purpose of the Study:

• To develop and evaluate hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) models for continuous pain intensity detection using EDA.
• To analyze various EDA features for enhanced objective pain assessment.
• To establish a non-invasive method for real-time pain monitoring.

Main Methods:

• Implementation of five parallel hybrid CNN-LSTM models.
• Analysis of EDA signals, including phasic components and sympathetic activity.
• Validation on the BioVid Heat Pain Dataset with 37 independent subjects.

Main Results:

• A parallel 1D CNN with stacked bi- and unidirectional LSTM achieved the best performance.
• The model demonstrated high accuracy in detecting four pain intensity levels (M-RMSE: 0.925, R-squared: 0.498).
• Robust performance was confirmed on independent subjects (M-RMSE: 0.957, R-squared: 0.495).

Conclusions:

• Hybrid CNN-LSTM architectures can analyze continuous EDA signals for objective pain intensity detection.
• This approach offers a reliable, non-invasive method for continuous pain monitoring.
• Findings support improved diagnosis, pain management, and patient outcomes in chronic pain.