Semi-Automatic Terminology Generation for Information Extraction from German Chest X-Ray Reports

Area of Science:

• Medical Informatics
• Natural Language Processing
• Radiology

Background:

• Structured data extraction from medical reports is crucial for research and care data warehousing.
• Radiology reports often use a telegraphic style, posing challenges for automated data extraction.

Purpose of the Study:

• To develop and evaluate a semi-automatic algorithm for generating a local terminology for information extraction from radiology reports.
• To map this local terminology to a standard ontology for enhanced data warehousing.

Main Methods:

• A two-phase algorithm for terminology matching was proposed.
• The first phase involves semi-automatic generation of a local terminology tailored to radiology report terms.
• The second phase maps the local terminology to a standard ontology.

Main Results:

• A local terminology with approximately 250 attributes and value patterns was created with 20 hours of radiologist input.
• The algorithm achieved an F1-Score of approximately 95% for information extraction in an evaluation with 100 chest X-ray reports.

Conclusions:

• The proposed semi-automatic approach efficiently generates local terminologies for radiology report analysis.
• This method significantly improves structured data extraction accuracy, supporting the development of medical data warehouses.