Angelou0516/BTXRD

---
license: cc-by-nc-nd-4.0
task_categories:
- image-segmentation
- object-detection
- image-classification
tags:
- medical-imaging
- x-ray
- radiograph
- bone
- tumor
- bone-tumor
pretty_name: BTXRD (Bone Tumor X-Ray Radiograph Dataset)
size_categories:
- 1K<n<10K
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

# 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