Radiology reports: a quantifiable and objective textual approach.

Area of Science:

• Radiology
• Natural Language Processing
• Machine Learning

Background:

• Radiology report quality is crucial for patient care.
• Objective methods for assessing radiology report quality are needed.
• Current methods for evaluating reporting style are often subjective.

Purpose of the Study:

• To assess the feasibility of using automated lexical analysis and machine learning for objective characterization of radiology reports.
• To develop a quality improvement tool for radiology reporting.

Main Methods:

• Quantified 12 lexical parameters from radiology reports of four radiologists.
• Included metrics like word usage, sentence count, readability, passive voice, and lexical complexity.
• Used a neural network with these parameters to identify report authorship.

Main Results:

• Observed significant differences in reporting characteristics among radiologists.
• The 12 lexical metrics accurately identified the author of 60 unknown reports.
• Demonstrated a robust parametric signature for individual reporting styles.

Conclusions:

• Automated, quantifiable methods can objectively analyze radiology reporting styles.
• These methods provide impartial feedback and detect deviations from group norms.
• The identified parameters are specific enough for author identification, aiding quality improvement and training.