Leveraging Fine-Tuned Large Language Models for Interpretable Pancreatic Cystic Lesion Feature Extraction and Risk Categorization

Key Points

• This research aims to utilize large language models to enhance feature extraction and risk categorization for pancreatic cystic lesions (PCLs).
• Developed fine-tuned large language models for parsing radiology reports.
• Created structured registries from extracted data to facilitate population-level research.
• Evaluated the models' performance in identifying and categorizing features of PCLs.
• Demonstrated improved accuracy in feature extraction from radiology reports using language models.
• Identified significant risk factors associated with PCLs, enhancing risk categorization capabilities.

Abstract