Indexing thoracic CT reports using a preliminary version of a standardized radiological lexicon (RadLex)

Area of Science:

• Radiology
• Medical Informatics
• Natural Language Processing

Background:

• Radiological reports contain critical information for patient care and research.
• Standardization of radiological terminology is essential for data analysis and interoperability.
• The Radiological Lexicon (RadLex) is a developing terminology standard for radiology.

Purpose of the Study:

• To validate a preliminary version of RadLex against terms in thoracic CT reports.
• To index the content of thoracic CT reports using RadLex term categories.
• To assess the coverage of RadLex for thoracic CT report terminology.

Main Methods:

• A text processor extracted terms from 200 thoracic CT reports.
• Extracted terms were matched against the RadLex database.
• Two radiologists manually indexed report content into predefined RadLex categories.
• Statistical analysis, including Kruskal-Wallis and Mann-Whitney tests, was performed.

Main Results:

• 84% of extracted terms were found in RadLex.
• An average of 16.2 encoded items and 3.2 findings were identified per report.
• Modifier, Anatomic Location, Relationship, and Finding were the most frequent categories.
• The frequency of distinct findings per report increased with patient age.

Conclusions:

• RadLex demonstrates substantial coverage of terms found in thoracic CT reports.
• Further development of RadLex applications is recommended for broader validation.
• Addressing automatic relationship encoding is crucial for future applications.