MedSG-Bench: A Benchmark for Medical Image Sequences Grounding

MedSG-Bench is the first benchmark for medical image sequences grounding, accepted as a Spotlight at NeurIPS 2025. The dataset is associated with MedSG-188K, a grounding instruction-tuning dataset, and the MedSeq-Grounder model. It was last updated on Hugging Face on October 24, 2025.

Use Cases

• Benchmarking medical image grounding models based on the provided sequences.
• Training instruction-following models for medical imaging based on the associated MedSG-188K dataset.
• Evaluating multimodal AI performance in healthcare contexts based on the benchmark's focus.

Strengths

• First benchmark for medical image sequences grounding.
• Accepted at NeurIPS 2025 Datasets and Benchmarks Track as a Spotlight.
• Associated with a 188K-instruction tuning dataset and a trained model.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license are unknown, which may limit suitability assessment.

Provenance

Source

MedSG-Bench

Collection Method

Likely created as a research benchmark for model evaluation.

Time Range

null

Freshness

Last updated 2025-10-24 05:25:18; freshness should be verified.

Geography

null