toolevalxm/MedVis-ChestXR

MedVis: Medical Vision Data Hub

📑 Paper    |    🌐 Project Page    |    💾 Released Resources    |    📦 Repo

This is the resource page of our MedVis resources collection on Huggingface.

Dataset

Dataset	Link
MedVis-ChestXR	🤗

Please also check the raw data after our processing if you are interested: [toolevalxm/MedVis-ChestXR-Raw](https://huggingface.co/datasets/toolevalxm/MedVis-ChestXR-Raw).

Models

Base Model / Training	MedVis		MedVis++
	Stage 1	Stage 2	Stage 1	Stage 2
LLaVA-Med 7B	🤗	🤗	🤗	🤗
BioMedCLIP	🤗	🤗	🤗	🤗

Introduction

While medical imaging datasets provide valuable visual information, generating accurate diagnostic descriptions requires expert knowledge. To address this, we adopt a vision-language model approach using GPT-4V for synthesizing detailed radiological reports and diagnostic annotations. GPT-4V provides state-of-the-art multimodal understanding capabilities at reasonable cost compared to domain-specific medical AI systems.

*Due to patient privacy requirements and institutional review board (IRB) protocols, we only release the ChestXR subset (this page) of the full MedVis dataset.

License

The license for this dataset is CC0.