Automated chest X-ray report generation with MAIRA-2, transforming a manual radiology workflow into a fast, AI-assisted drafting system for grounded and non-grounded findings.
Integrated multimodal clinical context, including frontal and lateral images, prior studies, and exam details, to produce more complete, verifiable reports and improve diagnostic relevance across real radiology use cases.
Delivered measurable performance gains, including over 70% RadFact Logical Precision, up to 90% Grounding Recall, and up to a 30% boost in BLEU-4 compared with earlier approaches.
Reduced reporting workload and errors by generating a strong first draft that was judged over 90% acceptable by expert review, making the tool a practical productivity boost for radiology teams.
Made with Bullet