Arcriser/

cxr-reasoning

	MedGemma LLaVA-Med 10K Reasoning Dataset




	
	Dataset Description

A curated collection of 10,000 medical image-reasoning pairs converted from the LLaVA-Med dataset into MedGemma's structured reasoning format. Each sample contains:

Original medical images (various modalities) Step-by-step diagnostic reasoning Final diagnoses with supporting evidence Clinically relevant web search terms

	Dataset Structure




	
	Example Instance

{ "image": <PIL.Image.Image… See the full description on the dataset page: https://huggingface.co/datasets/Arcriser/cxr-reasoning.