BoneLoss-PAN769: Expert-Annotated Panoramic Radiograph Dataset for Periodontal Bone-Loss Assessment

BoneLoss-PAN769 is an expert-annotated dataset of 769 panoramic dental radiographs with synchronized tooth-crown and alveolar-ridge annotations, stored in a single COCO-compatible JSON file. It extends the InReDD panoramic-radiograph collection and was built through a hybrid human–AI workflow: expert cementoenamel-junction (CEJ) and alveolar-ridge landmarks were combined with automated tooth instance segmentation to derive tooth-wise crown regions, which were then reviewed and validated by dental specialists.

Contents:
- 769 panoramic radiographs (patients aged 14–81; ~60% female / ~40% male), acquired on a Morita Veraviewepocs device.
- 18,809 expert-validated crown polygons and 769 alveolar-ridge polylines (plus maxillary/mandibular reference lines) — 21,064 annotations in total.
- COCO-compatible annotations extended with dental metadata (patient age, sex, tooth associations).

Intended uses: periodontal bone-loss quantification, tooth-crown instance segmentation, crown-to-root ratio analysis, and benchmarking of dental segmentation models.

This dataset accompanies the article "Descriptor: Expert-Annotated Panoramic Radiograph Dataset for Periodontal Bone-Loss Assessment (BoneLoss-PAN769)" (IEEE Data Descriptions). Reference implementation: https://github.com/thiagoavieira/boneloss-inference (archived at https://doi.org/10.5281/zenodo.20089016).

Ethics: data collection and annotation were approved by the institutional Research Ethics Committee (CAAE 51238021.2.0000.5419).