Mammo-CLIP

Mammo-CLIP: A Vision Language Foundation Model to Enhance Data Efficiency and Robustness in Mammography

A pip-installable package for zero-shot mammography analysis using the Mammo-CLIP model.

Installation

pip install mammoclip

Quick Start

Python API

import mammoclip

# Initialize model (downloads automatically)
model = mammoclip.MammoClipModel()

# Analyze an image (supports DICOM, PNG, JPEG, etc.)
results = model.predict("mammogram.dcm", {
    "mass": ["no mass", "mass"],
    "malignancy": ["no malignancy", "malignancy"],
    "density": ["fatty", "scattered areas of fibroglandular density", 
               "heterogeneously dense", "extremely dense"]
})

print(results)

Command Line Interface

# Basic usage
mammoclip-inference --image mammogram.dcm

# With custom prompts
mammoclip-inference --image mammogram.png --prompts custom_prompts.json

Supported Image Formats

• DICOM: .dcm, .dicom
• Standard Images: .png, .jpg, .jpeg, .tiff, .bmp

Model Information

This package is based on the research paper:

"Mammo-CLIP: A Vision Language Foundation Model to Enhance Data Efficiency and Robustness in Mammography"

• Paper: https://arxiv.org/abs/2409.03675
• Original Code: https://github.com/batmanlab/Mammo-CLIP

Citation

@article{shen2024mammoclip,
  title={Mammo-CLIP: A Vision Language Foundation Model to Enhance Data Efficiency and Robustness in Mammography},
  author={Shen, Shantanu and Xu, Haoyue and Weng, Jaden and Wu, Jay and Chen, Evangelia and Abbasi, Salma and Wang, Rayna and Bouzid, Hania and Rajpurkar, Pranav},
  journal={arXiv preprint arXiv:2409.03675},
  year={2024}
}

License

MIT License