Dental Structure Segmentation Dataset for Panoramic Radiographs (TL-pano)

This dataset is for non-commercial research purposes only.

The automatic detection of dental diseases using deep learning is an emerging field. The featureless nature of common dental diseases makes it difficult to detect and accurately stage using traditional detection methods and datasets. We propose an anonymous instance segmentation dataset, named TL-pano, for the segmentation of tooth and alveolar bone structures in panoramic radiographs, for use as a supporting dataset to improve the quality of dental caries and periodontal bone loss detection as distributed datasets. Our dataset contains 197 annotated and 114 non-annotated panoramic radiographs, with instance level segmentation annotations for tooth layers, tooth numbers and alveolar bone detection.

The primary classes of objects annotated in the radiographs include:

·         Composite - Class 0 – Any composite/reconstructive material present on the tooth eg. Fillings, Root Canals

·         Enamel – Class 1 - The enamel layer of the tooth

·         Pulp – Class 2 - The pulp layer of the tooth

·         Dentin – Class 3 - The dentin layer of the tooth

·         Tooth – Class 4 - The full outline of the tooth with attached FDI tooth numbering

·         Upper – Class 5 - The upper (Maxillary) alveolar bone

·         Lower – Class 6 - The lower (Mandible) alveolar bone

If the primary class is 'Tooth' (Class 4), the Quadrant classes include:

Upper Right - Class 0 - Upper right quadrant of teeth (FDI 1?)

Upper Left - Class 1 - Upper left quadrant of teeth (FDI 2?)

Lower Left - Class 2 - Lower left quadrant of teeth (FDI 3?)

Lower Right - Class 3 - Lower right quadrant of teeth (FDI 4?)

If the primary class is 'Tooth' (Class 4), the Tooth Number classes include:

Central Incisor - Class 0 - Central Incisor tooth (FDI ?1)

Lateral Incisor - Class 1 - Lateral Incisor tooth (FDI ?2)

Canine - Class 2 - Canine  tooth (FDI ?3)

First Premolar - Class 3 - First Premolar tooth (FDI ?4)

Second Premolar - Class 4 - Second Premolar tooth (FDI ?5)

First Molar - Class 5 - First Molar tooth (FDI ?6)

Second Molar - Class 6 - Second Molar tooth (FDI ?7)

Third Molar - Class 7 - Third Molar tooth (FDI ?8)

The dataset contains 197 "annotated" and 114 "unannotated" images.

"annotated" Images and annotations are stored in their own individual subfolders where a .JPG image has a corresponding .JSON annotation file exclusivly for that image.

.JSON annotations are stored as pixel value (x,y) coordinate polygons per object, with three classes "class_id" (0-6), "quadrant_id" (null, 0-3) and "tooth_type_id" (null, 0-7), where "quadrant_id" and "tooth_type_id" are null if "class_id" is not 4.

Annotations are processed to retain as much information as possible, so "class_id" 3, 4, 5, and 6 have overlap with other classes. Processing is required before using the dataset in a multiclass setting. When processing, the lower the class number the higher the priority of the object.

.JSON annotations are stored as:

Instance counts for each class are as follows: