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

Objective pain intensity assessment is critical to providing effective opioid-based treatment to suppress pain while preventing the development of addiction and tolerance. Previous studies have demonstrated the potential of electrodermal activity (EDA) for objective pain intensity detection using machine learning, but it is limited by intermittent detection (not continuous). In the present study, we implemented five parallel hybrid convolutional-long short-term memory (CNN-LSTM) models and assessed their respective efficacies in continuous pain intensity detection using EDA. Multiple EDA features, including phasic components and time-varying index of sympathetic activity, were analyzed. Among the five models, the parallel 1-dimentional CNN with a stacked bi- and unidirectional LSTM demonstrated the best performance in detecting four different pain intensity levels with an M-RMSE of 0.925 and an R-squared score of 0.498. The robustness of the model was tested with 37 independent subjects from the BioVid Heat Pain Dataset, achieving an M-RMSE of 0.957 and an R-squared score of 0.495. The findings of this study indicate that implementing parallel hybrid CNN-LSTM architectures to analyze continuous EDA signals have the potential to provide continuous and objective detection of pain intensity levels.Clinical Relevance- By providing a reliable and non-invasive measure of pain intensity, our findings will contribute to improved diagnosis, pain management strategies, and ultimately, better outcomes for chronic pain patients.