NEJM-Bench: 1,063 Medical Image Challenge Questions

1,063 multiple-choice visual questions drawn from the NEJM Image Challenge, a weekly clinical quiz published by the New England Journal of Medicine from October 2005 to April 2026. The benchmark covers a broad range of medical specialties and presents a medical image with five answer choices for identifying a diagnosis, finding, or underlying process. It was created by the author 'medicalsphere' and last updated on the Hugging Face platform in May 2026.

Use Cases

• Benchmarking visual question answering models based on medical images and multiple-choice questions.
• Training diagnostic AI assistants based on clinical image quizzes from a major medical journal.
• Evaluating model performance across medical specialties like radiology and dermatology based on the described coverage.
• Studying the intersection of computer vision and clinical reasoning based on the diagnosis-oriented question format.

Strengths

• 1,063 multiple-choice questions provide a substantial benchmark for evaluation.
• Questions span over 20 years of weekly publications from a prominent medical journal, suggesting temporal breadth.
• Covers a broad range of medical specialties including radiology, dermatology, and ophthalmology as mentioned in the description.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for large-scale training.
• Description metadata is limited; actual data quality and image accessibility require manual inspection after download.

Provenance

Source

New England Journal of Medicine (NEJM) Image Challenge

Collection Method

Likely compiled from published weekly quiz questions.

Time Range

October 2005 – April 2026

Freshness

Last updated 2026-05-04 22:28:06; freshness should be verified.