Diagnostic performance of artificial intelligence-aided caries detection on bitewing radiographs: a systematic review and meta-analysis

Abstract

The accuracy of artificial intelligence-aided (AI) caries diagnosis can vary considerably depending on numerous factors. This review aimed to assess the diagnostic accuracy of AI models for caries detection and classification on bitewing radiographs. Publications after 2010 were screened in five databases. A customized risk of bias (RoB) assessment tool was developed and applied to the 14 articles that met the inclusion criteria out of 935 references. Dataset sizes ranged from 112 to 3686 radiographs. While 86 % of the studies reported a model with an accuracy of ≥80 %, most exhibited unclear or high risk of bias. Three studies compared the model's diagnostic performance to dentists, in which the models consistently showed higher average sensitivity. Five studies were included in a bivariate diagnostic random-effects meta-analysis for overall caries detection. The diagnostic odds ratio was 55.8 (95 % CI= 28.8 - 108.3), and the summary sensitivity and specificity were 0.87 (0.76 - 0.94) and 0.89 (0.75 - 0.960), respectively. Independent meta-analyses for dentin and enamel caries detection were conducted and showed sensitivities of 0.84 (0.80 - 0.87) and 0.71 (0.66 - 0.75), respectively. Despite the promising diagnostic performance of AI models, the lack of high-quality, adequately reported, and externally validated studies highlight current challenges and future research needs.

Keywords: Adjunct methods; Assessment; Dental caries; Diagnostic techniques and procedures; Reference standards; Visual examination.

Fig. 1

Flow diagram illustrating the search and study selection process.

Fig. 2

Summary of the methodological quality assessment for the included studies. The red bar indicates high RoB, the yellow bar indicates unclear RoB and the green one indicates low RoB.

Fig. 3

Dot plot displaying the DOR of some of the networks in the included studies.

Fig. 4

Scatterplot showing the sensitivity and specificity results for some of the AI models reported in the included studies.

Fig. 5

Forest plot for the meta-analyses for overall caries detection, dentin caries detection, and enamel caries detection showing the summary sensitivity and specificity for the included studies with the confidence interval (95 % CI).

Fig. 6

HSROC curve showing the summary point for overall caries detection along with 95 % confidence region and 95 % prediction region estimates. The prediction region is the estimated 95 % probability range for the expected performance of future studies conducted similarly to those already analyzed.