PanoraDent: an annotated dental panoramic radiograph dataset with tooth-level findings

Description

PanoraDent contains 1,295 de-identified adult dental panoramic radiographs (OPGs) from two hospitals in Sylhet, Bangladesh: North East Medical College Hospital (425) and Mount Adora Hospital (870). It provides 8,204 bounding boxes in 11 classes, seven of which are target classes: missing tooth, dental caries, endodontically treated tooth, impacted third molar, retained root, periapical lesion and tooth fracture. Labels are given in YOLO and COCO formats. Each radiograph also has a tooth-level table (present and missing teeth per jaw, per-tooth counts of findings, FDI numbers of impacted and root-canal-treated teeth), derived by matching the human boxes to an automatic tooth-numbering detector; every row states its provenance and uncertain values are flagged. The release includes five stratified cross-validation folds (used for the benchmark in the accompanying Data in Brief article), a fixed 915/191/189 train/validation/test split, per-image metadata (hospital, size, annotation batch), a data dictionary, the annotation guideline and a change log. See README.md for the folder structure.

Steps to reproduce

Unzip, then train directly with Ultralytics: yolo detect train data=data.yaml model=yolo11m.pt imgsz=1280. For the cross-validated benchmark use splits/folds_5.csv. Column definitions are in structured_findings/data_dictionary.csv.

Institutions

• Shahjalal University of Science and Technology

  Sylhet Division, Sylhet
• North East Medical College

  Sylhet Division, Sylhet

Categories

Licence