Abstract
Practical pain assessment is essential in clinical practice. However, conventional methods for pain assessment depend primarily on patients’ self-reports, which are not feasible for non-communicative individuals, such as neonates and unconscious patients. Automated pain recognition presents a viable alternative that utilizes multimodal approaches integrating physiological signals and behavioral patterns. This survey examines recent advancements in Artificial Intelligence-based pain assessment techniques. We review current methodologies, challenges, and applications in healthcare. Furthermore, we discuss critical limitations, including dataset availability and model interpretability. Finally, we propose future research directions to enhance automated pain recognition systems’ accuracy and clinical integration.