wanglab/eurorad-gpt-oss-training-data

---
license: cc-by-4.0
task_categories:
- text-generation
- question-answering
language:
- en
tags:
- medical
- radiology
- diagnosis
- clinical-reasoning
- eurorad
- clinical-decision-support
size_categories:
- 1K<n<10K
---

<div align="center">

**Benchmarking and Adapting On-Device Large Language Models for Clinical Decision Support**

</div>

<div align="center">
<table align="center">
<tr>
<td><a href="[arXiv link - to be added]" target="_blank"><img src="https://img.shields.io/badge/arXiv-Paper-FF6B6B?style=for-the-badge&logo=arxiv&logoColor=white" alt="Paper"></a></td>
<td><a href="https://github.com/bowang-lab/on-device-LLM" target="_blank"><img src="https://img.shields.io/badge/GitHub-Code-181717?style=for-the-badge&logo=github&logoColor=white" alt="Code"></a></td>
<td><a href="https://huggingface.co/wanglab/on-device-LLM-gpt-oss-20b" target="_blank"><img src="https://img.shields.io/badge/HuggingFace-Model-FFBF00?style=for-the-badge&logo=huggingface&logoColor=white" alt="HuggingFace Model"></a></td>
<td><a href="https://huggingface.co/datasets/wanglab/eurorad-gpt-oss-training-data" target="_blank"><img src="https://img.shields.io/badge/HuggingFace-Dataset-28A745?style=for-the-badge&logo=huggingface&logoColor=white" alt="Dataset"></a></td>
</tr>
</table>
</div>

## Authors

<p align="center">
<a href="https://huggingface.co/alif-munim">Alif Munim</a>^{* 1},
<a href="https://scholar.google.com.hk/citations?hl=en&user=bW1UV4IAAAAJ">Jun Ma</a>^{* 1,2},
<a href="https://huggingface.co/omareng">Omar Ibrahim</a>^{* 1},
<b>Alhusain Abdalla</b>^{* 1},
Shuolin Yin³,
Leo Chen⁴,
<a href="https://scholar.google.ca/citations?user=37FDILIAAAAJ&hl=en">Bo Wang</a>^{† 1,5,6,7,8}
</p>

<p align="center">
^* Equal contribution &nbsp;&nbsp;&nbsp; ^† Corresponding author
</p>

<p align="center">
¹AI Collaborative Centre, University Health Network, Toronto, Canada<br>
²Princess Margaret Cancer Centre, University Health Network, Toronto, Canada<br>
³Department of Electrical and Computer Engineering, University of Toronto, Toronto, Canada<br>
⁴Division of Urology, Department of Surgery, St. Michael's Hospital, Unity Health Toronto and University of Toronto, Toronto, Canada<br>
⁵Peter Munk Cardiac Centre, University Health Network, Toronto, Canada<br>
⁶Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Canada<br>
⁷Department of Computer Science, University of Toronto, Toronto, Canada<br>
⁸Vector Institute for Artificial Intelligence, Toronto, Canada
</p>

## Highlights

- 1,894 EuroRad medical radiology cases with expert-level diagnostic reasoning
- Each case enhanced with detailed chain-of-thought reasoning generated by GPT-OSS 120B
- Used to fine-tune [wanglab/on-device-LLM-gpt-oss-20b](https://huggingface.co/wanglab/on-device-LLM-gpt-oss-20b) for on-device clinical decision support
- Part of a broader benchmark study on adapting on-device LLMs for medical tasks

## Dataset Overview

This dataset contains 1,894 medical radiology cases sourced from Eurorad, each enhanced with detailed step-by-step diagnostic reasoning generated by GPT-OSS 120B. It was used to train the LoRA fine-tuned model available at [wanglab/on-device-LLM-gpt-oss-20b](https://huggingface.co/wanglab/on-device-LLM-gpt-oss-20b).

Each case includes the clinical history, imaging findings, a differential diagnosis list, the confirmed final diagnosis, and a structured reasoning trace that maps symptoms to differentials and converges on the correct diagnosis.

## Dataset Structure

| Field | Type | Description |
|-------|------|-------------|
| `case_id` | string | Unique case identifier (e.g., case_0001) |
| `PostDescription` | string | Clinical history, patient presentation, and imaging findings |
| `DifferentialDiagnosisList` | string | List of differential diagnoses considered |
| `FinalDiagnosis` | string | Confirmed final diagnosis |
| `gptoss120b_reasoning` | string | Step-by-step diagnostic reasoning generated by GPT-OSS 120B |

## Dataset Statistics

| | |
|--|--|
| **Total Cases** | 1,894 |
| **Source** | Eurorad (European Society of Radiology) |
| **Reasoning Model** | GPT-OSS 120B |
| **File Format** | Parquet |
| **Language** | English |

## Usage

```python
from datasets import load_dataset

# Load the dataset
dataset = load_dataset("wanglab/eurorad-gpt-oss-training-data")

# Access training data
train_data = dataset['train']

# Example: print first case
print(train_data[0])
```

## Citation

> Citation will be updated upon arXiv submission and journal publication.

```bibtex
@article{munim2025ondevice,
title={Benchmarking and Adapting On-Device Large Language Models for Clinical Decision Support},
author={Munim, Alif and Ma, Jun and Ibrahim, Omar and Abdalla, Alhusain and Yin, Shuolin and Chen, Leo and Wang, Bo},
journal={},
year={2025}
}
```

## Limitations

- **Medical Validation Required**: Reasoning is AI-generated and requires clinical validation
- **Not for Clinical Use**: Dataset is for research and training purposes only
- **Specialty Coverage**: May not cover all radiology subspecialties equally
- **Reasoning Quality**: AI-generated reasoning may contain errors or hallucinations
- **Language**: English only

## Related Resources

- **Fine-tuned Model**: [wanglab/on-device-LLM-gpt-oss-20b](https://huggingface.co/wanglab/on-device-LLM-gpt-oss-20b)
- **Code Repository**: [github.com/bowang-lab/on-device-LLM](https://github.com/bowang-lab/on-device-LLM)

## Contact

For issues and questions, please open a discussion in this repository.
Corresponding author: Bo Wang — [e-mail]

> **Disclaimer**: This dataset is for research purposes only. The diagnostic reasoning is AI-generated and has not been clinically validated. Always consult qualified healthcare professionals for medical decisions.