Extracting diagnosis from Japanese radiological report

Area of Science:

• Natural Language Processing
• Medical Informatics
• Radiology

Background:

• Radiological reports contain crucial diagnostic information.
• Automated extraction of diagnostic assertions from Japanese medical text is challenging.
• Accurate interpretation of radiological findings requires understanding assertion polarity (positive/negative).

Purpose of the Study:

• To develop a method for extracting diagnoses with positive or negative assertions from Japanese radiological reports.
• To identify and utilize frequent verb patterns indicative of assertion polarity.
• To build a customized dictionary for improved term recognition.

Main Methods:

• A customized dictionary of 36,152 disease and radiological finding terms was created.
• Verb patterns indicating positive/negative assertions were identified and ranked by frequency.
• Rule-based extraction of (assertion, disease, verb pattern) pairs was performed on CT reports.
• The method was applied to 1,524 out of 5,000 Japanese CT reports.

Main Results:

• The study successfully extracted pairs of diagnostic assertions, diseases, and verb patterns.
• Frequent verb patterns were identified and used to formulate extraction rules.
• Initial rule-based extraction demonstrated feasibility for automated assertion identification.

Conclusions:

• The developed method shows promise for automatically extracting diagnostic assertions from Japanese radiological reports.
• Further refinement of rules and expansion of the dictionary can enhance extraction accuracy.
• This approach can aid in the efficient analysis of large volumes of medical text.