Application, performance and limitations of artificial intelligence for the diagnosis and prediction of oro-facial pain: a systematic review

Abstract

Background Oro-Facial Pain (OFP) disorders are a group of diseases characterized by overlapping clinical features leading to diagnostic and therapeutic challenges. Recent advances in artificial intelligence (AI) such as machine learning (ML) and deep learning (DL) have shown potential to improve the accuracy of diagnosis, prediction of pain and decision making in the clinic. The objective of this systematic review is to assess the use, diagnostic accuracy, and clinical utility of AI models in the diagnosis and prediction of OFP conditions. Methods This systematic review was performed following the PRISMA-DTA guidelines and registered at PROSPERO (CRD420261322606). A comprehensive electronic search was conducted for studies published from January 1, 2000 to February 1, 2026 in PubMed, Scopus, Embase, Cochrane Library, Web of Science and Google Scholar. Eligible studies examined AI-based methods for the diagnosis, classification, localization or prediction of OFP conditions. Independent reviewers performed study selection, data extraction, and quality assessment. Methodological quality was assessed using the QUADAS-2 tool and the certainty of the evidence was assessed using GRADE approach. Qualitative synthesis was performed due to substantial heterogeneity in study design, datasets, AI architectures and outcome measures. Results Twenty studies were included. These studies assessed AI applications for general diagnosis of orofacial pain, temporomandibular disorders, trigeminal neuralgia and facial pain syndromes, prediction of postoperative odontogenic pain, and localization of dental pain. The AI models were trained on heterogeneous data modalities including clinical records, questionnaires, thermography, radiographic imaging, MRI, neuroimaging and electronic health records. ML algorithms and artificial neural network (ANN) models demonstrated diagnostic accuracies ranging from 75% to 99%, demonstrating a potential for better performance on imaging-based and structured clinical datasets. Thermography-based ML models achieved the highest reported accuracy (99%) over other modalities, and MRI-based predictive models for temporomandibular disorders exhibited high discriminatory ability (AUC up to 0.899). Most studies were classified as low risk of bias in the patient selection and index test domains, but there were major concerns with regard to applicability and limited external validation in the reference standard domain. Overall certainty of evidence was rated as moderate. Conclusion AI-based systems have great potential as adjunctive tools for the diagnosis, classification and prediction of OFP conditions, especially when using structured clinical and imaging datasets. However, the current evidence is limited by methodological heterogeneity, retrospective study design, and lack of external validation. Future research should focus on prospective multi-center studies, standardized datasets, explainable AI frameworks, and the integration of multimodal AI approaches into clinically applicable decision support systems. Systematic Review Registration https://www.crd.york.ac.uk/PROSPERO/view/CRD420261322606, identifier CRD420261322606.