# MedText-Clinical-Processed

<p align="left">
    📑 <a href="https://arxiv.org/papers/xxxx.xxxxx" target="_blank">Paper</a> &nbsp&nbsp | &nbsp&nbsp 🌐 <a href="https://medtext.github.io/" target="_blank">Project Page</a> &nbsp&nbsp | &nbsp&nbsp 💾 <a href="https://huggingface.co/collections/toolevalxm/medtext-collection" target="_blank">Released Resources</a> &nbsp&nbsp | &nbsp&nbsp 📦 <a href="https://github.com/xmhtoolathlon/MedText-DataHub" target="_blank">Repo</a> 

This is the processed version of MedText-Clinical dataset, containing structured medical entities and relations extracted using BiomedBERT-Large.

**Dataset**
<table>
      <tr>
        <th>Dataset</th>
        <th>Link</th>
    </tr>
      <tr>
        <td>MedText-Clinical-Processed</td>
        <td style="background-color: #e6f3ff; text-align: center; vertical-align: middle;">
          <a href="https://huggingface.co/datasets/toolevalxm/MedText-Clinical-Processed">🤗</a>
        </td>
    </tr>
</table>

Please also check the raw data: [toolevalxm/MedText-Clinical-Raw](https://huggingface.co/datasets/toolevalxm/MedText-Clinical-Raw).

**Models Used**
- BiomedBERT-Large for entity extraction
- Clinical-Longformer for relation classification

**Introduction**

We use a pipeline approach combining BiomedBERT-Large for named entity recognition and Clinical-Longformer for relation extraction. The processing extracts medical concepts (diseases, medications, procedures) and their relationships from raw clinical notes.

*Due to licensing requirements, we only release the de-identified subset of the full dataset.