Machine Learning-Based Infant Cry Interpretation for Pain Detection and Distress Identification in Neonatal Intensive Care Units: A Scoping Review

Background: The interpretation of infant cries through artificial intelligence and machine learning presents significant opportunities for objective pain detection in neonatal intensive care units, yet implementation faces substantial challenges including limited clinical validation and data scarcity. This scoping review aimed to map and synthesize existing evidence regarding machine learning-based infant cry interpretation for pain detection and classification in NICU settings, examining methodological approaches, classification performance and implementation barriers. Methods: Following PRISMA-ScR guidelines, we searched PubMed/MEDLINE, Scopus, Google Scholar, Web of Science, and Iranian databases (SID, Magiran, IranDoc) for studies published between January ۲۰۲۰ and August ۲۰۲۶. Two independent reviewers screened ۴۷۶ articles, with ۲۵ studies meeting inclusion criteria. Quality assessment utilized ROBINS-I, RoB ۲, CASP and PROBAST. Findings were synthesized narratively due to study heterogeneity. Results: So far, machine learning techniques achieved high accuracy in classifying whether an infant was in pain or not and identifying many cry categories. Deep learning methods performed even better than traditional machine learning methods. Furthermore, multimodal approaches combining cry acoustics with facial expressions have also shown better results. Conclusion: Machine learning-based infant cry interpretation shows substantial promise for enhancing objective pain assessment in NICUs. Successful implementation requires validation in diverse populations, development of explainable AI methods, and integration into clinical workflows.

کلیدواژه ها:

Infant ، Pain-Detection ، Neonatal Intensive Care Unit ، Machine Learning ، Deep Learning ، Cry Classification