PET2Rep: PET/CT Image-Report Benchmark for Vision-Language Models

565 whole-body PET/CT image-report pairs comprise this benchmark dataset for evaluating vision-language models in radiology report generation. The dataset offers coverage of dozens of organs with metabolic information, reflecting real-world clinical PET imaging in oncology. It was authored by r1pples and last updated on November 14, 2025.

Use Cases

• Benchmark vision-language model performance for radiology report generation based on PET/CT image pairs.
• Train models to generate descriptive reports based on metabolic information across dozens of organs.
• Evaluate the clinical comprehensiveness of automated reports in oncologic PET imaging contexts.

Strengths

• Contains 565 paired PET/CT images and radiology reports.
• Covers dozens of organs with corresponding metabolic information.
• Designed specifically as a benchmark for vision-language model evaluation.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but other scale metrics like file formats and size are unknown.
• Freshness should be verified as the last update was November 14, 2025.

Provenance

Source

huggingface

Freshness

Last updated 2025-11-14 13:22:10