Multi-Level Analysis for Pain Recognition from Physiological Signals

This paper addresses the need for automatic pain recognition in healthcare without relying on expert feature extraction from physiological signals. Instead, it introduces a deep learning approach that combines classification and feature extraction methods by incorporating multi-level information for accurate pain discrimination. Experimental results are based on BioVid Heat Pain database and Emopain-2021 dataset, which demonstrate its superiority in high accuracy over conventional methods. Notably, in the classification of Pain-0 and Pain-4, the approach achieves average of 84.8% and 87.8% accuracies in cross-validation tests. This research harnesses the power of deep learning to improve pain recognition using physiological signals, eliminating the need for extensive medical expertise.