Deep Learning Detection of Proximal Caries on Pediatric Periapical Radiographs

DEIMv2-X achieved highest Macro-F1 0.538 & mAP50 0.486 for detecting interproximal caries on pediatric periapical radiographs (internal test); YOLOv11x had best recall 0.561. External validation favored YOLOv11x for Macro-F1 & precision. Incipient lesions remained challenging.

• Three architectures (Faster R-CNN, YOLOv11x, DEIMv2-X) were compared; detection performance was consistently higher for advanced than incipient radiographic lesions.
• External validation showed reduced performance across all models, with YOLOv11x achieving the highest overall Macro-F1 and precision.
• The study used 1,838 digital periapical radiographs from pediatric patients in mixed dentition, split into training (70%), validation (15%), and held-out internal test (15%) sets.

Automated summary

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