Clinical Scorecard: Evolution of Large Language Models in Radiology Structured Reporting: Historical Insights and Future Directions
At a Glance
Category
Detail
Condition
Radiology Structured Reporting (SR)
Key Mechanisms
Standardization of radiologic report content and use of IT tools including large language models (LLMs) to import, organize, and generate structured reports
Target Population
Radiologists and healthcare professionals involved in radiology reporting
Care Setting
Radiology departments and clinical radiology practice
Key Highlights
Structured reporting improves report quality, reduces errors, and enhances guideline compliance but may limit nuanced interpretations and requires significant resources.
Historical milestones include ACR communication guidelines (1991), RadLex lexicon (2006), and RSNA RadReport templates linking standardized vocabularies.
Large language models (LLMs), based on transformer architectures, represent a promising solution to integrate SR into radiologists’ workflows by automating and enhancing report generation.
Guideline-Based Recommendations
Diagnosis
Adopt standardized nomenclature and structured templates to improve clarity and consistency of radiology reports.
Management
Implement IT-based SR methods linking radiology vocabularies (RadLex, SNOMED CT, LOINC) to report elements.
Leverage LLMs to reduce manual effort and improve workflow integration in structured reporting.
Monitoring & Follow-up
Evaluate SR adoption rates and compliance with national and international guidelines.
Monitor radiologist engagement and feedback to optimize SR templates and LLM integration.
Risks
Potential rigidity of structured reports limiting nuanced clinical interpretation.
Resource-intensive creation and maintenance of SR templates.
Reluctance among radiologists to adopt SR due to workflow disruption.
Patient & Prescribing Data
Patients undergoing radiologic imaging requiring diagnostic interpretation and reporting
Structured reporting supported by LLMs can enhance report accuracy and accessibility, potentially improving diagnostic communication and patient management.
Clinical Best Practices
Use standardized vocabularies and ontologies (RadLex, SNOMED CT, LOINC) to ensure uniformity in reporting.
Incorporate LLMs to automate and streamline the generation of structured radiology reports.
Engage radiologists in the development and continuous updating of SR templates to ensure usability and clinical relevance.
Encourage institutional and policy-level incentives to promote SR adoption in routine clinical practice.
by Felix Busch, Lena Hoffmann, Daniel Pinto dos Santos, Marcus R. Makowski, Luca Saba, Philipp Prucker, Martin Hadamitzky, Nassir Navab, Jakob Nikolas Kather, Daniel Truhn, Renato Cuocolo, Lisa C. Adams, Keno K. Bressem
