Extracting actionable findings of appendicitis from radiology reports using natural language processing

Area of Science:

• Medical Informatics
• Natural Language Processing
• Radiology

Background:

• Radiology reports frequently contain critical findings requiring prompt clinical action.
• Automated detection of these actionable findings can enhance patient care by ensuring timely notification of referring physicians.
• This process also provides valuable feedback to radiologists regarding report outcomes.

Purpose of the Study:

• To develop and evaluate an automated method for identifying actionable findings related to appendicitis in radiology reports.
• To assess the system's ability to detect both direct and indirect indicators of appendicitis.

Main Methods:

• Utilized syntactic dependency patterns to identify individual assertions and related findings indicative of appendicitis.
• Aggregated information from all relevant statements within a report to determine the likelihood of appendicitis.
• Evaluated the method on a corpus of 400 radiology reports.

Main Results:

• The automated method achieved a precision of 91% for detecting actionable appendicitis findings.
• A recall of 83% was obtained, indicating effective identification of relevant reports.
• An overall F1-measure of 87% demonstrates the system's robust performance.

Conclusions:

• The developed method effectively automates the detection of actionable appendicitis findings in radiology reports.
• This approach holds significant potential for improving the efficiency and accuracy of clinical decision-making.
• Automated analysis of radiology reports can streamline communication and enhance patient management pathways.