New analysis epxlores the accuracy of AI in caries detection

Dental Tribune International

Thu. 9. January 2025

save

BUDAPEST, Hungary: Artificial intelligence (AI) is rapidly transforming the dental industry in a radical way, and its influence is being felt especially powerfully in diagnostics, where AI-powered platforms are capable of instantly and effectively analysing a wide range of visual information for a variety of dental conditions. To assess the accuracy of AI models in the difficult task of detecting approximal caries on bitewing radiographs, researchers at Semmelweis University in Budapest conducted a systematic review and meta-analysis. Their results confirm emerging understandings that AI models can be highly effective within a clinical setting but that the clinician remains of vital importance for final judgement and decision-making.

Twenty-one studies, of an initial 2,442, met the inclusion criteria. To be eligible, studies had to compare AI findings with expert consensus, and only randomised and non-randomised controlled trials were considered. The researchers employed statistical models to evaluate the diagnostic measures of sensitivity—the proportion of correctly identified teeth with caries—and specificity—the proportion of correctly identified healthy teeth. This method accounted for their interrelationship and combined results across studies to estimate overall performance.

The findings revealed AI’s impressive diagnostic accuracy. Pooled sensitivity was 94%, and specificity reached 91%, demonstrating the ability of AI to detect carious lesions and exclude healthy teeth effectively. The models showed high negative predictive values, suggesting AI’s reliability in identifying non-caries cases, though positive predictive values were moderate, indicating some potential for incorrectly identifying healthy teeth as having caries. The diagnostic odds ratio, which quantifies how much more likely a tooth with caries is to be flagged as decayed compared with a healthy tooth, was high, further underscoring AI’s robust performance, especially when compared with that of human experts, whose sensitivity has been found to typically lag at 24%–36% in other studies—although their specificity is high, at 94%–97%. In this meta-analysis, the diagnostic accuracy of AI could not be compared with that of experts because of a lack of data published in the included studies.

Clinically, the implications of using AI are significant. AI can offer substantial assistance in caries detection; however, the study emphasised the necessity of expert oversight to validate positive findings in order to prevent unnecessary treatments or missed diagnoses. By complementing human expertise, AI can streamline dental workflows, enhance diagnostic precision and reduce the effects of examiner variability.

The study, titled “Diagnostic accuracy of artificial intelligence for approximal caries on bitewing radiographs: A systematic review and meta-analysis”, was published in the December 2024 issue of the Journal of Dentistry.