Pain remains a major clinical challenge in the intensive care unit (ICU), especially in sedated, mechanically ventilated, or curarized patients due to their inability to self-report and the limited accuracy of behavioral tools. Therefore, innovative approaches must be developed. In this scenario, objective and observer-independent pain assessment can support and improve personalized analgesic management.

The aim of this review is to analyze the current artificial intelligence (AI) applications for automatic pain assessment (APA) in the ICU, focusing on the integration of biosignals, behavioral indicators, and multimodal data to detect nociceptive responses.

A systematic search was conducted in PubMed, Web of Science, and IEEE Xplore databases (2015–2025) using the terms pain assessment, critical care, artificial intelligence, machine learning, facial expression, pupillometry, heart rate variability, and nociception monitor. The scientific output was grouped into three main domains: behavioral and computer-vision methods, autonomic and electrophysiological indices, and multimodal and AI-driven integrated systems.

Conclusion. Although AI systems for APA in the ICU show promising performance, several challenges limit their clinical translation. Signal variability due to pharmacological, neurological, or hemodynamic factors may compromise model reliability. Moreover, the scarcity of labeled ICU datasets can affect generalizability. Ethical, regulatory, and interoperability issues should be addressed. Therefore, for routine implementation, large-scale validation across diverse ICU populations is required to confirm reliability, ensure fairness, and establish clinical utility.