ReXrank: Public Leaderboard for AI Radiology Report Generation
ReXrank, a public leaderboard and challenge for evaluating AI-powered radiology report generation, has been introduced. The platform, accessible at https://rexrank.ai, aims to standardize the assessment of AI models that generate radiology reports for chest X-rays. It incorporates ReXGradient, the largest test dataset comprising 10,000 studies, along with three public datasets: MIMIC-CXR, IU-Xray, and CheXpert Plus. ReXrank employs eight evaluation metrics and separately assesses models that generate only findings sections versus those providing both findings and impressions. This standardized framework enables meaningful comparisons of model performance and offers insights into robustness across diverse clinical settings. The announcement was made via arXiv (arXiv:2411.15122v2).
Key facts
- ReXrank is a public leaderboard and challenge for AI-powered radiology report generation.
- It is available at https://rexrank.ai.
- ReXrank incorporates ReXGradient, the largest test dataset with 10,000 studies.
- Three public datasets are used: MIMIC-CXR, IU-Xray, and CheXpert Plus.
- ReXrank employs 8 evaluation metrics.
- It separately assesses models generating only findings and those providing findings and impressions.
- The framework aims to standardize evaluation and enable meaningful comparisons.
- The announcement was made on arXiv with identifier arXiv:2411.15122v2.
Entities
Institutions
- ReXrank
- arXiv