README.md

---
license: apache-2.0
language:
- en
---

# MedM-VL-CT-3B-en

## Introduction

A medical LVLM, trained on **English** data, accepts text and **a single 3D CT volume** as input, and text-based results as output, enabling tasks such as **report generation** and **medical VQA**.

Here are the evaluation results on **M3D-Bench**:

<table>
  <tr>
    <td rowspan="2"> Method </td>
    <td align="center" colspan="4"> Report Generation </td>
    <td align="center" colspan="5"> Medical VQA </td>
  </tr>
  <tr align="center">
    <td> BLEU </td>
    <td> ROUGE </td>
    <td> METEOR </td>
    <td> BERT-Score </td>
    <td> Accuracy </td>
    <td> BLEU </td>
    <td> ROUGE </td>
    <td> METEOR </td>
    <td> BERT-Score </td>
  </tr>
  <tr>
    <td> RadFM </td>
    <td align="center"> 12.23 </td>
    <td align="center"> 16.49 </td>
    <td align="center"> 11.57 </td>
    <td align="center"> 87.93 </td>
    <td align="center"> 19.79 </td>
    <td align="center"> 16.39 </td>
    <td align="center"> 26.13 </td>
    <td align="center"> 21.33 </td>
    <td align="center"> 88.72 </td>
  </tr>
  <tr>
    <td> M3D-LaMed </td>
    <td align="center"> 15.15 </td>
    <td align="center"> 19.55 </td>
    <td align="center"> 14.38 </td>
    <td align="center"> 88.46 </td>
    <td align="center"> 75.78 </td>
    <td align="center"> 49.38 </td>
    <td align="center"> 52.39 </td>
    <td align="center"> 33.58 </td>
    <td align="center"> 91.53 </td>
  </tr>
  <tr>
    <td> MedM-VL-CT-3B-en </td>
    <td align="center"> <b>49.81</b> </td>
    <td align="center"> <b>52.45</b> </td>
    <td align="center"> <b>49.27</b> </td>
    <td align="center"> <b>90.38</b> </td>
    <td align="center"> <b>80.12</b> </td>
    <td align="center"> <b>56.56</b> </td>
    <td align="center"> <b>59.96</b> </td>
    <td align="center"> <b>39.75</b> </td>
    <td align="center"> <b>92.85</b> </td>
  </tr>
</table>


## Quickstart

Please refer to [MedM-VL](https://github.com/MSIIP/MedM-VL).