Google MedGemma WebUI

This project is a web interface built with Gradio to run image-to-text inference using Google's medgemma-4b-it vision-language model. MedGemma is a collection of Gemma 3 variants that are trained for performance on medical text and image comprehension.

Features

• Upload and analyze medical images (X-ray, MRI, etc.) (accepts ".jpg" ".png", but not ".dicom" raw images. ".dicom" images needs to first be converted to normal image formats)
• Ask free-form medical questions related to the image
• Powered by medgemma-4b-it, a multimodal transformer
• Clean, interactive, and easy to understand web UI using Gradio

Requirements

• Needs to run while not a root user
• Python 3.10+
• Download the medgemma-4b-it model from Hugging Face:
  • download the project from github, and navigate into its folder

    git clone https://github.com/Google-Health/medgemma.git
    cd medgemma

  • download the model file (requires huggingface account)

    huggingface-cli download google/medgemma-4b-it --local-dir checkpoints/medgemma-4b-it

Install Python library dependencies:

pip install torch transformers accelerate bitsandbytes gradio pillow

Usage

• Run the app

  python medgemma.py

• The program will prompt it is serving on "0.0.0.0:7860" , you will need to use a web browser to visit "server-IP:7860"

• Upload a medical image (e.g., X-ray.jpg).

• Ask a question like:

  “What is shown in this image?”

  “Is there evidence of pneumonia?”

• Click Submit.

• The model will respond with an expert-level interpretation in the right window