Artificial intelligence for pain assessment via facial expression recognition (2015–2025): a systematic review

Abstract

Background: Although accurate pain assessment is crucial in clinical care, pain evaluation is traditionally based on self-report or observer-based scales. Artificial intelligence (AI) applied to facial expression recognition is promising for objective, automated, and real-time pain assessment. Methods: The study followed PRISMA guidelines. We searched PubMed/MEDLINE, Scopus, Web of Science, Cochrane Library, and the IEEE Xplore databases for the literature published between 2015 and 2025 on the applications of AI for pain assessment via facial expression analysis. Eligible studies included original articles in English applying different AI techniques. Exclusion criteria were neonatal/pediatric populations, non-facial approaches, reviews, case reports, letters, and editorials. Methodological quality was assessed using the RoB 2 tool (for RCTs) and adapted appraisal criteria for AI development studies. This systematic review was registered in PROSPERO (https://doi.org/10.17605/OSF.IO/N9PZA). Results: A total of 25 studies met the inclusion criteria. Sample sizes ranged from small experimental datasets (n < 30) to larger clinical datasets (n > 500). AI strategies included machine learning models, convolutional neural networks (CNNs), recurrent neural networks such as long short-term memory (LSTM), transformers, and multimodal fusion models. The accuracy in pain detection varied between ~70% and > 90%, with higher performance observed in deep learning and multimodal frameworks. The risk of bias was overall moderate, with frequent concerns related to small datasets and lack of external validation. No meta-analysis was performed due to heterogeneity in datasets, methodologies, and outcome measures. Discussion: AI-based facial expression recognition shows promising accuracy for automated pain assessment, particularly in controlled settings and binary classification tasks. However, evidence remains limited by small sample sizes, methodological heterogeneity, and scarce external validation. Large-scale multicenter studies are required to confirm clinical applicability and to strengthen the certainty of evidence for use in diverse patient populations.