Kan AI estimere osteoporoserisiko ud fra rutine tandrøntgenbilleder af kæbeknogle ?

Osteoporose påvirker ofte kæbeknoglemineraltætheden, før der opstår systemiske symptomer. AI trænet på tandrøntgenbilleder kunne estimere knoglemineraltæthed uden yderligere stråling. Dette kunne muliggøre opportunistisk screening under tandlægebesøg. Nøjagtigheden afhænger af billedkvalitet og kalibrering på tværs af forskellige billedsystemer.

Background

Osteoporosis often affects jaw bone density before causing systemic symptoms, making opportunistic screening during dental visits attractive. Deep-learning models trained on panoramic dental radiographs (orthopantomograms) analyze trabecular bone microarchitecture to estimate systemic bone loss. Reported performance in validation cohorts reaches sensitivities around 80–90% for identifying low bone mineral density, approaching the accuracy of dual-energy X-ray absorptiometry (DEXA) scans. Variability in X-ray equipment, the absence of standardized acquisition and calibration protocols, and the need for broader validation across diverse populations currently limit clinical adoption. Current tools remain largely research-oriented, though several commercial dental AI platforms have begun to integrate osteoporosis risk-assessment features. AI training relies on large annotated datasets linking radiographic jaw features to DEXA-derived bone mineral density or clinical osteoporosis diagnoses, with cross-site validation essential to ensure generalizability. Calibration across different panoramic systems and patient subgroups is critical to reduce false positives and negatives. Future directions include federated learning to harmonize multi-vendor datasets and integration of AI outputs into electronic health records to facilitate clinician follow-up.