MedGemma LLaVA-Med 10K: Medical Image-Reasoning Pairs

A collection of 10,000 medical image-reasoning pairs converted from the LLaVA-Med dataset into MedGemma's structured reasoning format. The dataset was created by author Manusinhh and last updated on June 29, 2025. Each sample contains an original medical image, step-by-step diagnostic reasoning, a final diagnosis with supporting evidence, and clinically relevant web search terms.

Use Cases

• Training vision-language models for medical diagnosis based on the step-by-step reasoning pairs.
• Benchmarking AI-generated clinical reasoning against the provided final diagnoses and evidence.
• Developing search tools for medical literature based on the included clinically relevant web search terms.
• Fine-tuning models for structured medical report generation based on the image-reasoning format.

Strengths

• Contains 10,000 curated medical image-reasoning pairs.
• Provides structured reasoning with step-by-step diagnostics and final diagnoses.
• Includes clinically relevant web search terms for each sample.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known but other scale details like file size and formats are unknown.
• Data may reflect biases inherent to the original LLaVA-Med source dataset.

Provenance

Source

Converted from the LLaVA-Med dataset.

Collection Method

Curated conversion into MedGemma's structured reasoning format.

Time Range

null

Freshness

Last updated 2025-06-29 06:19:16.

Geography

null