Kan AI opdage visse sygdomme ved at se på billeder af tænder ?

AI kan allerede hjælpe med at opdage visse tandtilstande ved at analysere røntgenbilleder såsom panoramiske røntgenbilleder og cone-beam computertomografi (CBCT) scanninger. Konvolutionelle neurale netværk (CNN'er) trænet på mærkede tandrøntgenbilleder har vist ydeevne sammenlignelig med menneskelige eksperter i at identificere problemer som huller i tænderne, parodontose og tandcaries, hvor nogle undersøgelser rapporterer nøjagtigheder over 90% under kontrollerede forhold. Dog forbliver generalisering på tværs af forskellige befolkningsgrupper, billedudstyr og kliniske protokoller udfordrende, og disse værktøjer bruges typisk som beslutningsstøttesystemer snarere end selvstændige diagnostiske løsninger. Udvidet klinisk validering og regulatorisk godkendelse er undervejs i mange jurisdiktioner.

— Beriget 13. maj 2026 · Kilde: American Dental Association — https://www.ada.org/resources/research/science-and-research-institute/ada-seal-of-acceptance

Background

AI-based dental diagnostics rely primarily on radiographic and photographic image analysis. Convolutional neural networks (CNNs) trained on labeled dental radiographs have achieved expert-level performance in detecting cavities, periodontal disease, dental caries, and other pathologies, with several studies reporting accuracies above 90% in controlled settings (American Dental Association, 2026). The U.S. National Institute of Dental and Craniofacial Research (NIDCR, 2026) similarly notes that AI systems have demonstrated high accuracy in identifying tooth decay, gum disease, and oral cancer from radiographic and intraoral images.

Key technical and clinical challenges include generalization across diverse patient populations, imaging equipment variability, and differences in clinical imaging protocols. Current systems are therefore positioned as decision-support tools rather than standalone diagnostic solutions (American Dental Association, 2026). Broader clinical validation and regulatory approval remain active areas of research and development in multiple jurisdictions. Performance is also influenced by image quality and the specific machine-learning algorithms employed (NIDCR, 2026).