Dataset for DentScanNet

This dataset accompanies the paper “Deep Learning–Based Real-Time Annotation of Periodontal Ultrasound with Clinical Indices”. It is the first publicly available annotated dataset for full-mouth periodontal high-frequency ultrasound imaging, covering all tooth types and both buccal and lingual surfaces, including posterior teeth.

Cohorts

• UCSD (healthy): 20 adult volunteers (12 female, 8 male; age 28.1 ± 6.5 years) with no clinical signs of periodontal disease, recruited under UC San Diego IRB approval (IRB-170912).
• USC (diseased): 12 subjects (7 female, 5 male; age 49.2 ± 15.2 years) with stage II–IV periodontitis, recruited at the Herman Ostrow School of Dentistry of the University of Southern California under a separate IRB protocol.
• All data were de-identified using study identifiers prior to release.

Annotations: Each image is paired with up to six binary segmentation masks across two task types:

• Point landmarks: Gingival Margin (GM), Cementoenamel Junction (CEJ), Alveolar Bone Crest (ABC).
• Anatomical regions: Tooth structure (TOOTH), gingival soft tissue (GINGIVA), alveolar bone (BONE).