Radiology Report Evaluation

The field develops ways to score report-generation and report-review systems for clinical accuracy, completeness, and harmful errors—not just similarity to a reference report. Approaches include finding-level extraction and matching, clinical graphs, question answering and entailment, severity-aware LLM judges, and direct radiologist review. Benchmarks span chest X-rays, CT, PET, MRI, and longitudinal reports, with growing attention to spatial grounding, report-writing conventions, and whether findings are supported by images. A central tension is that fluent reports and high text-overlap scores can conceal omissions or unsupported claims, while more clinically aligned metrics still need to be transparent, consistent, and validated across settings.