Real-Time Pain Assessment from Electrodermal Activity Using Deep Learning

Area of Science:

• Biomedical Engineering
• Neuroscience
• Machine Learning

Background:

• Objective pain assessment is challenging due to subjective self-reports.
• Electrodermal activity (EDA) shows promise for indicating pain-related autonomic responses.
• Current deep learning models for biosignals can be computationally intensive for real-time applications.

Purpose of the Study:

• To develop an efficient fully convolutional network (FCN) for automated pain recognition using EDA signals.
• To enable real-time pain monitoring with low computational complexity for wearable systems.

Main Methods:

• A fully convolutional network (FCN) was designed to analyze temporal patterns in EDA signals.
• The model was trained and evaluated on the AI4Pain dataset for three-class pain classification (No Pain, Low Pain, High Pain).
• Real-time inference performance was assessed for latency and accuracy.

Main Results:

• The proposed FCN achieved 79.23% accuracy in offline pain classification.
• Real-time operation demonstrated a latency of 0.47 ms with 73.14% accuracy.
• Convolutional architectures offer a balance between performance and computational efficiency.

Conclusions:

• The FCN model provides an effective and computationally efficient approach for physiological pain monitoring.
• This study supports the development of real-time pain detection systems using wearable sensors and EDA.
• Deep learning with EDA signals presents a viable pathway for objective pain assessment.