Clinically annotated dataset boosts deep learning detection of dental restorations on panoramic X-rays

A deep learning model (Faster R-CNN) trained on 2,434 panoramic radiographs achieved an [email protected] of 0.75 for detecting dental restorations/appliances; small-target recall improved from 0.59 to 0.81. Implants had highest precision (0.93). Braces precision was 0.32 due to artifacts…

• Performance varied widely: implant precision 0.93, braces precision 0.32 due to overlapping artifacts.
• Processing speed was under 1.2 s per image on a standard laptop.
• The study lacked external multicenter validation, limiting generalizability to other clinical settings.

Automated summary

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