GatorTron

GatorTron is an open-source large language model (LLM) developed specifically for the healthcare domain, trained on over 90 billion words of clinical text, including millions of electronic health records (EHRs) from the University of Florida Health system.

Key Capabilities

• Clinical Entity Recognition: Highly tuned out-of-the-box for extracting complex medical concepts, medications, and procedures from unstructured clinical notes.
• Relation Extraction: Can accurately identify relationships between medical entities (e.g., linking a specific dosage to a specific medication span).
• Semantic Textual Similarity: Can determine if two distinct clinical narratives refer to the same underlying medical condition or patient state.

Usage in Healthcare

GatorTron is designed to be a foundational model that researchers and health systems can fine-tune for their specific use cases. Unlike proprietary API-based models, GatorTron can be downloaded and run entirely on-premises, which is a critical requirement for hospitals handling sensitive Protected Health Information (PHI) under HIPAA regulations.