Style-Aware Radiology Report Generation with RadGraph and Few-Shot Prompting
This addresses the need for clinically accurate and personalized radiology reports for radiologists, though it is incremental as it builds on existing methods with a novel style adaptation focus.
The paper tackled the problem of conflating content and style in radiology report generation by proposing a two-step approach that extracts content from images and verbalizes it to match specific radiologist styles using RadGraph and LLMs, resulting in AI-generated reports that were rated as indistinguishably tailored to individual styles in human evaluations.
Automatically generated reports from medical images promise to improve the workflow of radiologists. Existing methods consider an image-to-report modeling task by directly generating a fully-fledged report from an image. However, this conflates the content of the report (e.g., findings and their attributes) with its style (e.g., format and choice of words), which can lead to clinically inaccurate reports. To address this, we propose a two-step approach for radiology report generation. First, we extract the content from an image; then, we verbalize the extracted content into a report that matches the style of a specific radiologist. For this, we leverage RadGraph -- a graph representation of reports -- together with large language models (LLMs). In our quantitative evaluations, we find that our approach leads to beneficial performance. Our human evaluation with clinical raters highlights that the AI-generated reports are indistinguishably tailored to the style of individual radiologist despite leveraging only a few examples as context.