BTXRD — Bone Tumor X-Ray Radiograph Dataset

Radiographs of bones for classification, localization, and segmentation of primary bone tumors.

Details

Property	Value
Modality	X-ray (2D radiograph)
Anatomy	Upper/lower limbs, pelvis
Total images	3,746
Normal	1,879
Benign tumor	1,525
Malignant tumor	342
License	CC BY-NC-ND 4.0

Fields

• image — X-ray radiograph (RGB)
• mask — segmentation mask (L, pixel values 0-9, see class map below)
• image_id — e.g. IMG000001.jpeg
• bboxes_json — JSON string; list of {label, class_id, bbox: [x1,y1,x2,y2]}
• Metadata columns from original dataset.xlsx: center, age, gender, body-part flags (hand, ulna, radius, ...), pathology flags (tumor, benign, malignant), tumor-type flags, region flags (upper limb, lower limb, pelvis), view flags (frontal, lateral, oblique)

Mask class map

ID	Class
0	background
1	osteochondroma
2	multiple osteochondromas
3	simple bone cyst
4	giant cell tumor
5	osteofibroma
6	synovial osteochondroma
7	other bt (benign tumor)
8	osteosarcoma
9	other mt (malignant tumor)

Normal images have an all-zero mask. Masks are rasterized from the original LabelMe polygon annotations. Bounding boxes (one per lesion, paired with each polygon in the source data) are preserved separately in bboxes_json.

Splits

The original release does not define fixed train/val/test splits; the paper uses an 80/20 random split. All 3,746 samples are provided under a single train split — consumers should generate their own splits as needed.

Citation

Yao, S., Huang, Y., Wang, X., et al. A Radiograph Dataset for the Classification, Localization, and Segmentation of Primary Bone Tumors. Scientific Data 12, 88 (2025). DOI: 10.1038/s41597-024-04311-y

Source

• Figshare: https://figshare.com/articles/dataset/27865398
• License: CC BY-NC-ND 4.0