MedTrinity-25M: 25 Million Multimodal Medical Records with Multigranular Annotations

MedTrinity-25M consists of 25 million multimodal medical records featuring multigranular annotations, developed by UCSC-VLAA for ICLR 2025. The dataset provides large-scale image-text pairings designed to advance the training and evaluation of medical Multimodal Large Language Models (MLLMs).

Use Cases

• Training medical MLLMs using multigranular text annotations
• Fine-tuning vision-language models on 25 million medical image-text pairs
• Benchmarking zero-shot medical image classification via textual descriptions

Strengths

• 25 million record count
• Multigranular annotation levels
• ICLR 2025 peer-reviewed provenance

Limitations

• High storage and computational requirements for 25 million multimodal records
• Lack of explicit column-level documentation in the primary repository summary

Provenance

Source

UCSC-VLAA, ICLR 2025 paper 'MedTrinity-25M: A Large-scale Multimodal Dataset with Multigranular Annotations for Medicine'

Freshness

Current as of 2025 publication.