Tag
The paper introduces a method using corpus characterization and inverse constitutional fine-tuning to improve the stylistic alignment of AI-generated radiology reports with authentic radiologist writing. This approach achieves significant gains in text alignment metrics, demonstrating effectiveness for style-aware report generation.