Related Experiment Video
Updated: Jul 24, 2025
09:16
Multi-Modal Signals for Analyzing Pain Responses to Thermal and Electrical Stimuli
Published on: April 5, 2019
10.8K
Design and Evaluation of Deep Learning Models for Continuous Acute Pain Detection Based on Phasic Electrodermal
IEEE Journal of Biomedical and Health Informatics
|July 3, 2023
Summary
Researchers developed a deep learning model using electrodermal activity (EDA) to objectively assess pain continuously. This approach accurately detects pain onset and levels, offering a potential alternative to subjective pain scales.
Area of Science:
- Biomedical Engineering
- Neuroscience
- Machine Learning
Background:
- Current pain assessment is subjective, relying on self-reported scales, leading to potential issues with medication dosage and opioid addiction.
- Electrodermal activity (EDA) has shown promise as an objective physiological signal for pain detection.
- Existing machine learning and deep learning methods have limitations in continuously detecting acute pain and its onset from EDA signals.
Purpose of the Study:
- To evaluate deep learning models for continuous acute pain detection using electrodermal activity (EDA) signals.
- To accurately detect the onset of pain using physiological markers derived from EDA.
- To develop an objective pain assessment method to aid in precise medication prescription.
Main Methods:
- Evaluated 1D-CNN, LSTM, and hybrid CNN-LSTM architectures for continuous pain detection from phasic EDA features.
- Utilized a dataset of 36 healthy volunteers subjected to thermal grill-induced pain stimuli.
- Extracted and analyzed the time-frequency spectrum of phasic EDA (TFS-phEDA) as a key physiomarker.
Main Results:
- A parallel hybrid architecture combining temporal convolutional neural networks (TCN) and stacked bidirectional/unidirectional LSTMs achieved the best performance.
- The optimal model demonstrated a 77.8% F1-score for continuous pain detection within 15-second intervals.
- Validation on the BioVid Heat Pain Database showed 91.5% accuracy in recognizing elevated pain levels compared to baseline.
Conclusions:
- Deep learning models, particularly hybrid architectures, show significant feasibility for continuous pain detection using EDA.
- The TFS-phEDA feature is a highly effective physiomarker for distinguishing pain states.
- This objective approach could lead to more accurate pain management and reduced opioid dependency.
