LLMAIx

LLMAIx extracts predefined entities from unstructured medical text using privacy-preserving large language models to convert clinical free text into structured data for clinical research.

Key Features:

• Privacy-Preserving Local Execution: LLMAIx integrates on local hospital hardware without transferring patient data to external servers.
• Structured Information Extraction: The pipeline extracts predefined entities from unstructured clinical text, including symptoms, laterality, and TNM staging information.
• Anonymization Capability: LLMAIx performs anonymization of clinical letters as part of its information extraction workflow.
• Four-Step Processing Protocol: The workflow comprises problem definition and data preparation, data preprocessing, LLM-based information extraction, and output evaluation.

Scientific Applications:

• Clinical Letter Anonymization: LLMAIx anonymizes clinical letters from patients with pulmonary embolism to protect patient privacy while preserving research utility.
• Symptom and Laterality Extraction: The pipeline extracts symptoms and pulmonary embolism laterality from clinical letters to support clinical characterization.
• TNM Stage Extraction from Pathology Reports: LLMAIx extracts TNM staging information from pathology reports, as demonstrated on 100 pathology reports from The Cancer Genome Atlas Program (TCGA).
• Large-Scale Clinical Data Analysis: By converting unstructured clinical text into structured data, LLMAIx facilitates quantitative investigations and clinical decision-making.

Methodology:

The LLMAIx protocol consists of four main processing steps: problem definition and data preparation, data preprocessing, LLM-based information extraction, and output evaluation.