Uncertainty-Aware Ordinal Deep Learning for Automated Periapical Index Assessment on Dental Radiographs

This dataset accompanies the PAI-SegScore software repository for uncertainty-aware ordinal deep learning applied to automated Periapical Index assessment on dental radiographs. The study is built on the DenPAR PAI v1 cohort, which is derived from the DenPAR: Annotated Intra-Oral Periapical Radiographs Dataset for machine learning. The parent DenPAR dataset contains intraoral periapical radiographs with alveolar crestal bone level annotations, keypoints such as CEJ and apex, tooth bounding boxes, tooth identifiers, and tooth segmentation masks, and is intended for dental imaging research on alveolar bone-loss detection. PAI-SegScore adds endodontist expert-verified Periapical Index labels to a single-cohort subset, producing a frozen training, validation, and locked-test split for class-agnostic apex detection, apical crop extraction, ordinal grading, and uncertainty-aware analysis. The repository includes the study code and supporting artifacts for the DenPAR PAI workflow and is intended for research use in clinician-in-the-loop endodontic decision support, requiring clinical correlation and prospective validation before deployment. DenPAR Zenodo record: 10.5281/zenodo.16645076 DenPAR article DOI: 10.1038/s41597-025-05906-9