Deep learning caries detection on pediatric bitewings outperforms general dentists

On pediatric bitewing radiographs, a deep learning model detected caries with higher recall (0.51 vs 0.29) and F1-score (0.44 vs 0.29) than general dentists. The model underestimated depth, while dentists overestimated. (retrospective, no external validation)

• Diagnostic accuracy study using 247 test bitewing radiographs of primary teeth; reference standard from two experienced dentists using ICCMS classification.
• Model recall 0.51 vs dentists 0.29; precision 0.41 vs 0.31; F1 0.44 vs 0.29; mAP 0.41 vs 0.22. No CIs or p-values reported.
• Retrospective, single-dataset evaluation; no external or prospective validation; reference standard not histological; number of general dentists not specified.

RadPigeon summaries are original and for information only. They are not clinical advice.