---
dataset_info:
  features:
  - name: image_id
    dtype: int64
  - name: date
    dtype: string
  - name: question
    dtype: string
  - name: relevant_context
    dtype: string
  - name: option_A_votes
    dtype: string
  - name: option_B_votes
    dtype: string
  - name: option_C_votes
    dtype: string
  - name: option_D_votes
    dtype: string
  - name: option_E_votes
    dtype: string
  - name: total_number_votes
    dtype: string
  - name: brier_score
    dtype: float64
  - name: image
    dtype: image
  - name: image_hash
    dtype: string
  - name: options
    sequence: string
  - name: answer
    dtype: string
  splits:
  - name: test
    num_bytes: 69947825.0
    num_examples: 947
  download_size: 69646346
  dataset_size: 69947825.0
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
---

# NEJM - New England Journal of Medicine Image Challenge

## Description
This dataset contains clinical cases from the NEJM Image Challenge series. Each case presents clinical images and patient information, requiring diagnosis selection from multiple options. We greatly appreciate and build from the original data source available at https://github.com/cx0/nejm-image-challenge. We modify the format slightly to have `question`, `options`, and `answer` fields as described below:

## Data Fields
- `question`: Clinical case description with patient presentation
- `options`: Multiple choice diagnostic options
- `answer`: The correct diagnosis
- `image`: Clinical image(s) associated with the case

## Splits
- `test`: Test data for evaluation

## Usage
```python
from datasets import load_dataset

dataset = load_dataset("OctoMed/NEJM-Image-Challenge")
```

## Citation

If you find our work helpful, feel free to give us a cite!

```
@article{ossowski2025octomed,
  title={OctoMed: Data Recipes for State-of-the-Art Multimodal Medical Reasoning},
  author={Ossowski, Timothy and Zhang, Sheng and Liu, Qianchu and Qin, Guanghui and Tan, Reuben and Naumann, Tristan and Hu, Junjie and Poon, Hoifung},
  journal={arXiv preprint arXiv:2511.23269},
  year={2025}
}
```