A Natural Language Processing and Machine Learning Approach to Identification of Incidental Radiology Findings in...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Radiology Reporting

Background:

• Incidental findings in emergency department (ED) diagnostic imaging are common and may indicate serious conditions.
• Manual review of radiology reports for these findings can be challenging in busy clinical settings.
• Automating the identification of incidental findings is crucial for timely patient care.

Purpose of the Study:

• To develop and validate a supervised machine learning model.
• To utilize natural language processing (NLP) for automated recognition of incidental findings.
• To analyze radiology reports of patients discharged from the ED.

Main Methods:

• Retrospective analysis of computed tomography (CT) reports from discharged trauma patients.
• Manual labeling of CT reports by two independent annotators.
• Development and validation of a random forest model using NLP and regular expressions.

Main Results:

• The random forest model achieved high performance with an area under the curve of 0.92 in the validation set.
• The model demonstrated a sensitivity of 92.2%, specificity of 79.4%, and negative predictive value of 97.2%.
• Strong performance was consistent across various ED settings, including trauma centers and community EDs.

Conclusions:

• Machine learning and NLP can effectively classify incidental findings in ED CT reports.
• The model offers high sensitivity and negative predictive value, suggesting its utility in clinical practice.
• Automating report review can improve the identification of incidental findings and facilitate timely interventions.