Radiology Report Generation
Overview
Radiology report generation studies how to turn images such as chest X-rays, CT scans, and MRI into clinical findings and impressions. Common approaches use vision-language models with image-text alignment, anatomical or disease-specific features, retrieval of similar cases, and structured clinical concepts; newer systems also use agents, reinforcement learning, and prior studies to guide or refine reports. The field has improved on language and clinical-label benchmarks, but fluent text can still omit findings, invent details, or miss their location and change over time. A central challenge is evaluating factual accuracy and usefulness across hospitals and modalities, since text similarity alone does not reliably measure clinical correctness.
Latest papers
159 papers20 papers in the last four weeks, up 5% on the four weeks before. 0.1% of all new papers.