New benchmark CRS-Bench evaluates medical image encoder reliability

Researchers have developed CRS-Bench, a new benchmark designed to evaluate the reliability of medical image encoders. Unlike previous methods that focused solely on discrimination, CRS-Bench assesses encoders across four dimensions: discrimination, calibration, label efficiency, and robustness. The benchmark utilizes datasets from dermatology, ophthalmology, and radiology, including a CheXpert-to-MIMIC-CXR shift to simulate institutional changes. The Clinical Reliability Score (CRS) is introduced as a composite metric that combines these dimensions, offering a more comprehensive approach to selecting medical image encoders than traditional AUROC scores alone. AI

影响 Provides a more robust framework for selecting medical image encoders, potentially improving diagnostic accuracy and efficiency in healthcare.

排序理由 The item is a research paper introducing a new benchmark for evaluating AI models. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

• APTOS 2019
• CheXpert
• CRS-Bench
• dermatology
• ISIC 2019
• MedGemma
• medical image encoders
• MedSigLIP
• MIMIC-CXR
• ophthalmology
• radiology

• 论文
• 其他

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →