Deep learning model accurately classifies horizontal bone loss in mandibular first molars on CBCT

A deep learning model achieved 0.829 overall accuracy for classifying horizontal bone loss on cone-beam CT of mandibular first molars, comparable to a dentist's 0.820 accuracy.

• Overall diagnostic accuracy was 0.829 versus a dentist’s 0.820 (no significant difference).
• For the most severe bone loss (degree III), the model achieved a precision of 0.906 and an F1-score of 0.866.
• The model was trained and tested on 505 CBCT images with expert annotations.

Automated summary

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