Framework for Extracting Critical Findings in Radiology Reports

Area of Science:

• Radiology
• Medical Informatics
• Healthcare Compliance

Background:

• Critical results reporting guidelines mandate timely communication of urgent findings.
• Current oversight processes for critical results reporting can be manual and time-consuming.
• Ensuring compliance with reporting timelines is essential for patient safety and quality care.

Purpose of the Study:

• To develop and evaluate a generic report processing pipeline for automated extraction of critical findings from radiology reports.
• To enable automated quality and compliance oversight of critical results reporting.
• To analyze the prevalence and types of critical findings in a large dataset of radiology exams.

Main Methods:

• A generic report processing pipeline was designed to identify critical findings within dictated radiology reports.
• The pipeline was applied to a production dataset of 1,210,858 radiology exams.
• Algorithm performance was assessed using an annotated dataset of 327 sentences, achieving 91.4% accuracy (95% CI 87.6-94.2%).

Main Results:

• The automated pipeline successfully extracted critical findings with high accuracy.
• Critical findings were most frequently diagnosed in Computed Tomography (CT) and Magnetic Resonance (MR) exams.
• Intracranial hemorrhage and fluid collection were the most common critical findings identified.
• Approximately 1.6% of the analyzed exams contained at least one of the ten focused critical findings.

Conclusions:

• Automated extraction of critical findings from radiology reports is feasible and accurate.
• This methodology facilitates efficient quality assurance and compliance monitoring for critical results reporting.
• The findings support the use of such systems for research, workflow optimization, and regulatory adherence.