Impact of labelling inaccuracy and image noise on tooth segmentation in panoramic radiographs using federated, centralized, and local learning
FL matches or exceeds CL and outperforms LL across corruption scenarios while preserving privacy. Per-client loss trajectories provide an effective anomaly-detection mechanism and support FL as a practical, privacy-preserving approach for scalable clinical AI deployment.
Authors
- Lukas Esterle (ORCID: https://orcid.org/0000-0002-0248-1552)
- Alexandros Iosifidis (ORCID: https://orcid.org/0000-0003-4807-1345)
- Ruben Pauwels (ORCID: https://orcid.org/0000-0002-9462-7546)
- Khuram Naveed (ORCID: https://orcid.org/0000-0001-8286-6139)
- Sanyam Jain (ORCID: https://orcid.org/0000-0002-6851-3246)
- Johan Andreas Balle Rubak
Institutions
- Aarhus University (DK)
- Aarhus University Hospital (DK)
- Tampere University (FI)
Publication Details
- Journal
- Dentomaxillofacial Radiology
- Published
- 2026-01-16
- DOI
- https://doi.org/10.1093/dmfr/twag001
- Citations
- 1
- Primary Topic
- Dental Radiography and Imaging
- Type
- article
- Field-Weighted Citation Impact
- 0.00