MedEBench: 1,200+ Medical Operation Images for Text-Guided Editing

MedEBench is a benchmark for text-guided image editing in the medical domain. It contains over 1,200 real-world medical operation samples spanning 13 anatomical regions and 70 clinically relevant editing operations. The dataset was created by LIUMinghao and was last updated on June 4, 2025.

Use Cases

• Benchmarking text-to-image editing models based on the described 70 clinical editing operations.
• Training medical image synthesis models using the provided original and ground truth image pairs.
• Developing region-of-interest (ROI) aware editing tools based on the included mask data.
• Evaluating the clinical realism of generated medical images across 13 anatomical regions.

Strengths

• Contains over 1,200 real-world medical operation samples.
• Covers 13 anatomical regions, providing anatomical diversity.
• Defines 70 clinically relevant editing operations for structured evaluation.

Limitations

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

Provenance

Source

LIUMinghao on Hugging Face.

Collection Method

Likely collected from real-world medical operation samples, but the specific gathering method is not detailed.

Freshness

Last updated 2025-06-04 11:39:37; freshness should be verified.