Related Experiment Video
Updated: Jun 12, 2026
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
AI-assisted extraction of opportunistic findings from oncologic CT reports using large language models
Kevin Roitero1, Luigi Nardone2, Davide Lacitignola3
1University of Udine, Udine, Italy. kevin.roitero@uniud.it.
Scientific Reports
|June 10, 2026
Summary
Opportunistic imaging findings are often missed in cancer scans. An AI system using Large Language Models (LLMs) can effectively extract these secondary findings from radiology reports, improving patient care.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Opportunistic findings at imaging (iOFs) are common but often underreported in oncologic computed tomography (CT) examinations.
- Factors contributing to underreporting include radiologist workload, lack of awareness, and report structure.
Purpose of the Study:
- To investigate the detection and reporting of iOFs in routine oncologic CT examinations.
- To develop and validate an AI-based system for extracting iOFs and linking them to management recommendations.
Main Methods:
- A multi-phase study involving 156 oncology patients and five radiologists.
- Assessment of inter-reader agreement for ten predefined iOFs.
- Development of a Large Language Model (LLM) system to extract iOFs from free-text reports.
Main Results:
- High concordance was observed in iOF detection across radiologists.
- The LLM system demonstrated high agreement with expert annotations (F1=0.995 internal, 0.980 validation).
- Previously undocumented iOFs requiring follow-up were identified in 18.6% of patients.
Conclusions:
- LLM-based analysis can systematically extract documented opportunistic findings from radiology reports.
- This approach supports structured communication and may improve clinical workflows and patient management.
- AI tools can enhance the utilization of incidental findings in oncologic imaging.
