README.md

---
language:
- en
pretty_name: RAM-H1200
size_categories:
- 1K<n<10K
task_categories:
- image-segmentation
- image-classification
task_ids:
- semantic-segmentation
- instance-segmentation
- multi-class-classification
tags:
- medical
- radiography
- x-ray
- rheumatoid-arthritis
- musculoskeletal
- svdh
- bone-segmentation
- joint-localization
- bone-erosion
- jsn
license: cc-by-4.0
---

# RAM-H1200

## Dataset Summary

RAM-H1200 is a multi-task full-hand radiograph dataset for rheumatoid arthritis (RA) related image analysis. It is designed to support several clinically relevant computer vision tasks, including:

- hand bone structure segmentation
- bone erosion related segmentation
- joint localization for Sharp/van der Heijde (SvdH) scoring
- joint-level SvdH bone erosion (BE) scoring
- joint-level SvdH joint space narrowing (JSN) scoring

The dataset contains full-hand radiographs in BMP format, COCO-format annotations for segmentation and joint detection, joint-level ROI crops for scoring tasks, and study-level metadata.

## Homepage

- Dataset repository: `https://huggingface.co/datasets/TokyoTechMagicYang/RAM-H1200-v1`
- Benchmark repository: `https://github.com/YSongxiao/RAM-H1200`

## License

This dataset is released under the **Creative Commons Attribution 4.0 International (CC BY 4.0)** license.

## Supported Tasks and Applications

RAM-H1200 supports the following research tasks:

- **Segmentation**
  - Bone segmentation on full-hand radiographs
  - Bone erosion related segmentation

- **Detection / Localization**
  - Joint localization for BE-related regions
  - Joint localization for JSN-related regions

- **Classification / Scoring**
  - Joint-level SvdH BE score prediction
  - Joint-level SvdH JSN score prediction

Potential use cases include:

- automated RA severity assessment
- multi-task medical image analysis
- musculoskeletal structure segmentation
- joint-level radiographic scoring
- benchmarking AI systems for RA-related radiograph analysis

## Dataset Structure

```text
RAM-H1200-v1/
|-- Segmentation/
|   |-- train/
|   |   |-- JP_HMCRD_P0001_20210203_6791_L.bmp
|   |   |-- JP_HMCRD_P0001_20210203_6791_R.bmp
|   |   |-- ...
|   |   |-- _annotations_bone_rle.coco.json
|   |   |-- _annotations_be_rle.coco.json
|   |-- val/
|   |   |-- ...
|   |   |-- _annotations_bone_rle.coco.json
|   |   |-- _annotations_be_rle.coco.json
|   |-- test/
|   |   |-- ...
|   |   |-- _annotations_bone_rle.coco.json
|   |   |-- _annotations_be_rle.coco.json
|-- SvdH_Scoring/
|   |-- SvdH_BE_Scoring/
|   |   |-- train/
|   |   |   |-- JP_HMCRD_P0001_20210203_6791_L/
|   |   |   |   |-- CMC-T.bmp
|   |   |   |   |-- IP.bmp
|   |   |   |   |-- L.bmp
|   |   |   |   |-- MCP-I.bmp
|   |   |   |   |-- ...
|   |   |   |-- _annotations_be_joint_detection.coco.json
|   |   |   |-- _annotation_be_scores.json
|   |   |-- val/
|   |   |   |-- ...
|   |   |   |-- _annotations_be_joint_detection.coco.json
|   |   |   |-- _annotation_be_scores.json
|   |   |-- test/
|   |   |   |-- ...
|   |   |   |-- _annotations_be_joint_detection.coco.json
|   |   |   |-- _annotation_be_scores.json
|   |-- SvdH_JSN_Scoring/
|   |   |-- train/
|   |   |   |-- JP_HMCRD_P0001_20210203_6791_L/
|   |   |   |   |-- CMC-M.bmp
|   |   |   |   |-- CMC-R.bmp
|   |   |   |   |-- CMC-S.bmp
|   |   |   |   |-- MCP-I.bmp
|   |   |   |   |-- ...
|   |   |   |-- _annotations_jsn_joint_detection.coco.json
|   |   |   |-- _annotation_jsn_scores.json
|   |   |-- val/
|   |   |   |-- ...
|   |   |   |-- _annotations_jsn_joint_detection.coco.json
|   |   |   |-- _annotation_jsn_scores.json
|   |   |-- test/
|   |   |   |-- ...
|   |   |   |-- _annotations_jsn_joint_detection.coco.json
|   |   |   |-- _annotation_jsn_scores.json
|-- Metadata.xlsx
```

## Data Organization

### 1. Segmentation

The `Segmentation/` directory contains full-hand radiographs in BMP format, organized into `train`, `val`, and `test` splits.

A typical filename looks like:

```text
JP_HMCRD_P0001_20210203_6791_L.bmp
```

This naming scheme generally encodes:

- country or source prefix
- acquisition center
- anonymized patient identifier
- study date (de-identified via a consistent temporal offset per patient)
- image identifier
- hand side (`L` for left, `R` for right)

Each split contains two COCO-format annotation files:

- `_annotations_bone_rle.coco.json`
- `_annotations_be_rle.coco.json`

#### Bone Segmentation Annotations

`_annotations_bone_rle.coco.json` stores segmentation masks using COCO RLE encoding. The annotation categories include anatomical structures such as:

- Capitate
- Hamate
- Lunate
- Scaphoid
- Trapezium
- Trapezoid
- Radius
- Ulna
- MC1--MC5
- PP1--PP5
- DP1--DP5

The annotation file also contains some additional categories related to non-bony structures or acquisition artifacts, such as soft tissue or implants.

Example COCO annotation:

```json
{
  "id": 1,
  "image_id": 0,
  "category_id": 30,
  "bbox": [14.0, 198.0, 852.0, 1233.0],
  "area": 515212.0,
  "segmentation": {
    "size": [1431, 893],
    "counts": "..."
  }
}
```

#### Bone Erosion Related Segmentation Annotations

`_annotations_be_rle.coco.json` provides segmentation annotations related to bone erosion patterns. The category set includes:

- `Non-SvdH-BE`
- `SvdH-BE-50`
- `SvdH-BE-90`

These annotations are also stored in COCO RLE format.

### 2. SvdH BE Scoring

The `SvdH_Scoring/SvdH_BE_Scoring/` directory contains ROI crops for bone erosion scoring. Each case is stored in a separate folder named by a case identifier.

Example:

```text
JP_HMCRD_P0001_20210203_6791_L/
```

A typical BE case folder contains 16 ROI images corresponding to joints or anatomical regions such as:

- `CMC-T.bmp`
- `IP.bmp`
- `L.bmp`
- `Tm.bmp`
- `R.bmp`
- `U.bmp`
- `MCP-T.bmp`
- `MCP-I.bmp`
- `MCP-M.bmp`
- `MCP-R.bmp`
- `MCP-S.bmp`
- `PIP-I.bmp`
- `PIP-M.bmp`
- `PIP-R.bmp`
- `PIP-S.bmp`

Each split also includes:

- `_annotations_be_joint_detection.coco.json`
- `_annotation_be_scores.json`

#### BE Joint Detection

`_annotations_be_joint_detection.coco.json` stores joint localization annotations in COCO format. The categories map to BE-relevant joints or regions, including:

- `R`
- `U`
- `L`
- `CMC-T`
- `S`
- `Tm`
- `IP`
- `MCP-T`
- `MCP-I`
- `MCP-M`
- `MCP-R`
- `MCP-S`
- `PIP-I`
- `PIP-M`
- `PIP-R`
- `PIP-S`

#### BE Score Labels

`_annotation_be_scores.json` stores ground-truth joint-level BE scores indexed by full image filename.

Example:

```json
{
  "JP_HMCRD_P0167_20111230_3497_L.bmp": {
    "BE_MCP-T": 0,
    "BE_MCP-I": 1,
    "BE_MCP-M": 0,
    "BE_MCP-R": 0,
    "BE_MCP-S": 0,
    "BE_IP": 0,
    "BE_PIP-I": 0,
    "BE_PIP-M": 0,
    "BE_PIP-R": 1,
    "BE_PIP-S": 1,
    "BE_CMC-T": 0,
    "BE_Tm": 1,
    "BE_S": 0,
    "BE_L": 0,
    "BE_U": 0,
    "BE_R": 0
  }
}
```

### 3. SvdH JSN Scoring

The `SvdH_Scoring/SvdH_JSN_Scoring/` directory contains ROI crops for joint space narrowing scoring.

A typical JSN case folder contains 15 ROI images corresponding to:

- `CMC-M.bmp`
- `CMC-R.bmp`
- `CMC-S.bmp`
- `SC.bmp`
- `SR.bmp`
- `STT.bmp`
- `MCP-T.bmp`
- `MCP-I.bmp`
- `MCP-M.bmp`
- `MCP-R.bmp`
- `MCP-S.bmp`
- `PIP-I.bmp`
- `PIP-M.bmp`
- `PIP-R.bmp`
- `PIP-S.bmp`

Each split also includes:

- `_annotations_jsn_joint_detection.coco.json`
- `_annotation_jsn_scores.json`

#### JSN Joint Detection

`_annotations_jsn_joint_detection.coco.json` stores COCO-format joint localization annotations. Categories include:

- `CMC-M`
- `CMC-R`
- `CMC-S`
- `SC`
- `SR`
- `STT`
- `MCP-T`
- `MCP-I`
- `MCP-M`
- `MCP-R`
- `MCP-S`
- `PIP-I`
- `PIP-M`
- `PIP-R`
- `PIP-S`

#### JSN Score Labels

`_annotation_jsn_scores.json` stores ground-truth joint-level JSN scores indexed by full image filename.

Example:

```json
{
  "JP_HMCRD_P0167_20111230_3497_L.bmp": {
    "JSN_MCP-T": 2,
    "JSN_MCP-I": 0,
    "JSN_MCP-M": 0,
    "JSN_MCP-R": 0,
    "JSN_MCP-S": 0,
    "JSN_PIP-I": 0,
    "JSN_PIP-M": 0,
    "JSN_PIP-R": 0,
    "JSN_PIP-S": 0,
    "JSN_STT": 0,
    "JSN_SC": 0,
    "JSN_SR": 0,
    "JSN_CMC-M": 0,
    "JSN_CMC-R": 0,
    "JSN_CMC-S": 0
  }
}
```

## Metadata

The file `Metadata.xlsx` contains study-level metadata. Key columns include:

- `Mapped Image Stem`
- `StudyID`
- `Normalized PatientID`
- `isRA`
- `Sex`
- `Age`
- `Center`
- `PixelSpacing`
- `ImageSize`
- `LR`

These fields provide normalized identifiers, demographic information, acquisition center information, study date, image geometry, and hand laterality.

## Splits

RAM-H1200 is distributed with predefined splits:

- `train`
- `val`
- `test`

These splits are consistently provided for:

- segmentation
- BE scoring
- JSN scoring

## Data Loading Notes

This repository stores raw files rather than a single tabular annotation file. Depending on the task, users will typically load data as follows:

- use BMP images together with the corresponding COCO JSON files for segmentation or detection tasks
- use per-case ROI folders together with score JSON files for BE and JSN scoring tasks
- use `Metadata.xlsx` for study-level metadata lookup and cohort analysis

## Example Usage

### Load COCO annotations

```python
import json
from pathlib import Path

ann_path = Path("Segmentation/train/_annotations_bone_rle.coco.json")
with ann_path.open("r", encoding="utf-8") as f:
    coco = json.load(f)

print(len(coco["images"]))
print(len(coco["annotations"]))
print(coco["categories"][:5])
```

### Load BE score labels

```python
import json
from pathlib import Path

label_path = Path("SvdH_Scoring/SvdH_BE_Scoring/train/_annotation_be_scores.json")
with label_path.open("r", encoding="utf-8") as f:
    labels = json.load(f)

sample_key = next(iter(labels))
print(sample_key)
print(labels[sample_key])
```

### Load JSN score labels

```python
import json
from pathlib import Path

label_path = Path("SvdH_Scoring/SvdH_JSN_Scoring/train/_annotation_jsn_scores.json")
with label_path.open("r", encoding="utf-8") as f:
    labels = json.load(f)

sample_key = next(iter(labels))
print(sample_key)
print(labels[sample_key])
```

## Intended Uses

RAM-H1200 is intended for research and benchmarking in:

- rheumatoid arthritis radiograph analysis
- automated scoring of structural damage
- medical image segmentation
- joint localization and ROI extraction
- multi-task learning with hand radiographs

## Out-of-Scope Uses

This dataset is not intended for:

- direct clinical deployment without independent validation
- standalone medical decision-making
- patient re-identification
- non-research use without checking the dataset license and ethics approvals

## Source Data

RAM-H1200 consists of anonymized full-hand radiographs and derived annotations from multiple acquisition centers. It includes full-image labels, ROI-level labels, and metadata relevant to RA-related structural assessment.

## Personal and Sensitive Information

The dataset uses anonymized patient and study identifiers. Metadata is limited to research-relevant study and demographic information and does not include direct personal identifiers.

## Bias, Risks, and Limitations

- The dataset may reflect center-specific acquisition protocols and patient populations.
- Annotation quality depends on the consistency of expert labeling and task definitions.
- Some anatomical regions or score levels may be imbalanced.
- Models trained on this dataset may not generalize to other institutions, scanners, or populations without additional validation.
- The dataset is intended for research use, not for direct clinical diagnosis or treatment planning.

## Citation

If you use RAM-H1200 in your research, please cite the dataset and the associated paper.

### BibTeX

```bibtex
@misc{yang2026ramh1200,
  title={RAM-H1200: A Unified Evaluation and Dataset on Hand Radiographs for Rheumatoid Arthritis},
  author={Songxiao Yang and Haolin Wang and Yao Fu and Junmu Peng and Lin Fan and Hongruixuan Chen and Jian Song and Masayuki Ikebe and Shinya Takamaeda-Yamazaki and Masatoshi Okutomi and Tamotsu Kamishima and Yafei Ou},
  year={2026},
  eprint={2605.05616},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2605.05616}
}
```

## Suggested Citation

If you use the benchmark code or experimental settings, we also recommend citing:

```bibtex
@article{yang2026ram,
  title={RAM-W600: A Multi-Task Wrist Dataset and Benchmark for Rheumatoid Arthritis},
  author={Yang, Songxiao and Wang, Haolin and Fu, Yao and Tian, Ye and Kamishima, Tamotsu and Ikebe, Masayuki and Ou, Yafei and Okutomi, Masatoshi},
  journal={Advances in Neural Information Processing Systems},
  volume={38},
  year={2026}
}
```

## Acknowledgements

We thank the annotators, clinicians, and collaborating institutions who contributed to the collection, curation, and quality control of RAM-H1200.

## Contact

For questions, issues, or collaboration inquiries, please contact:

- `Songxiao Yang, Yafei Ou`
- `syang(at)ok.sc.e.titech.ac.jp, yafei.ou(at)riken.jp`
- `https://yafeiou.github.io/RAM10K`