Abstract
Automated tooth detection and numbering in dental radio- graphs is a key component of computer-aided dental diagnosis. Modern multi-class detectors assign each detected tooth one of 32 FDI classes but make predictions independently, often producing duplicate tooth num- bers that can lead to inconsistent dental charts. Existing approaches address duplicates through post-processing heuristics without gradient feedback or through more complex set-prediction architectures. We pro- pose a soft duplication loss that enforces the anatomical constraint that each tooth number should be predicted at most once. Class probabili- ties are summed across detections for each FDI class, and excess prob- ability mass above one is penalized during training. The loss operates within standard one-stage detectors, requires no architectural modifica- tion, and adds no inference overhead. Across two public panoramic ra- diograph datasets and multiple inference resolutions, the proposed loss improves detection performance and reduces duplicate predictions, with best observed gains of 2.6 points in mAP50, 2.3 points in F1, and 26% fewer duplicates. It is complementary to post-processing-based per-class deduplication, and their combination achieves the highest mean F1 on both datasets. These results show that a simple differentiable anatomical constraint can improve tooth detection and numbering while preserving efficient one-stage inference.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/ODIN_028.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/ODIN_028_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=6Bfx1Ldgiq
BibTex
@InProceedings{DelNoa_AnatomyConstrained_MICCAISAT2026,
author = { Delbari, Noam AND Hirschhorn, Ariel AND Specktor-Fadida, Bella},
title = { { Anatomy-Constrained Duplication Loss for Tooth Detection in Panoramic Radiographs } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17270},
month = {pending},
page = {pending}
}