This repository is a curated survey of research applying CLIP (Contrastive Language-Image Pre-training) models to medical imaging tasks. It serves as a comprehensive resource for researchers and practitioners interested in leveraging vision-language models for medical image analysis, classification, segmentation, and report generation. The survey aims to provide an organized overview of datasets, pre-training strategies, and CLIP-driven applications in the medical domain.

How It Works

The repository categorizes existing work based on the application of CLIP in medical imaging, including pre-training methodologies (multi-scale, data-efficient, knowledge-enhanced), and downstream tasks like classification, dense prediction, and cross-modal retrieval. It highlights various datasets used for training and evaluation, such as ROCO, MIMIC-CXR, and PadChest, and lists specific CLIP variants like PubMedCLIP and BioViL.

Quick Start & Requirements

This repository is a survey and does not have a direct installation or execution command. It lists research papers with links to their respective code and datasets where available.

Highlighted Details

• Comprehensive taxonomy of CLIP applications in medical imaging.
• Extensive list of relevant datasets with domain, size, and source information.
• Categorization of pre-training strategies and CLIP-driven applications.
• Links to numerous research papers, many with associated code repositories.

Maintenance & Community

The repository is maintained by zhaozh10 and is associated with a paper accepted by Medical Image Analysis. Further community interaction details are not provided in the README.

Licensing & Compatibility

The repository itself does not specify a license. Individual research papers linked within may have their own licenses.

Limitations & Caveats

This is a survey of existing research, not a runnable codebase. Users must refer to individual linked papers for specific implementation details, dependencies, and licensing. The rapid pace of research means the survey may not be exhaustive of the very latest publications.