Artificial Intelligence in Early Childhood Caries Detection and Prediction: A Systematic Review and Meta-Analysis

Purpose: To conduct a systematic review of artificial intelligence (AI) in aiding clinicians with the prediction and detection specifically for early childhood caries (ECC). Methods: A search was performed across PubMed^®, EMBASE^®, Scopus, Web of Science^TM, IEEE, and grey literature (Google Scholar) databases. The Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) tool was utilized to evaluate the potential bias. Statistical analyses were performed using random effects models in STATA 17.0, with Metandi and MIDAS packages. Results: Of 350 identified studies, 21 met the inclusion criteria. Nine studies demonstrated a low risk of bias in all QUADAS-2 categories. Studies were divided into two cate- gories: detection of ECC (7 studies) and prediction of ECC (14 studies). For ECC detection, studies reported accuracy between 78 to 86 percent, sen- sitivity ranged from 67 to 96 percent, and specificity ranged from 81 to 99 percent. In prediction-focused research, accuracies varied widely from 60 to 100 percent, sensitivity from 20 to 100 percent, and specificity from 54 to 94 percent. Six studies with adequate data were selected for the meta-analysis. The combined sensitivity of these studies was calculated at 80 percent (95 percent confidence interval [95% CI] equals 0.84 to 0.95), while the combined specificity reached 81 percent (95% CI equals 0.67 to 0.90). The diagnostic odds ratios ranged from 4 to 83, with a median of 17. Conclusion: AIs diagnostic accuracy for early childhood caries matches that of experienced dentists, emphasizing AI's potential to significantly aid in the detection and prediction of ECC.