Related Experiment Videos
Automated computer-assisted categorization of radiology reports
Bijoy J Thomas1, Hugue Ouellette, Elkan F Halpern
1Division of Musculoskeletal Radiology, Department of Radiology, Massachusetts General Hospital, 32 Fruit St., YAW 6E, Boston, MA 02114, USA.
AJR. American Journal of Roentgenology
|January 27, 2005
Summary
An automated method accurately categorizes radiography reports into normal, fracture, or neither categories. This validated computerized approach offers high sensitivity and specificity across multiple sites and radiologists.
Area of Science:
- Radiology Informatics
- Medical Imaging Analysis
- Computational Pathology
Background:
- Radiograph report categorization is crucial for clinical analysis and research.
- Manual review of narrative text reports is time-consuming and prone to variability.
- Automated methods can improve efficiency and consistency in report analysis.
Purpose of the Study:
- To develop and validate an automated computerized method for categorizing narrative text radiograph reports.
- To assess the accuracy and reproducibility of the automated method across different sites and radiologists.
Main Methods:
- A text search algorithm using Boolean logic was created to classify reports.
- The algorithm was refined using 512 ankle radiography reports.
- Validation was performed on 750 spine and extremity reports from three sites, interpreted by 44 radiologists.
Main Results:
- The computerized classification achieved high accuracy: normal (91.6% specificity, 91.3% sensitivity), neither normal nor fracture (87.8% sensitivity, 94.9% specificity), and fracture (94.1% sensitivity, 98.1% specificity).
- Performance metrics showed no significant differences across the three clinical imaging sites (p >0.05).
Conclusions:
- Automated categorization of narrative radiography reports is highly sensitive and specific.
- The method is effective across different imaging sites and radiologists.
- This tool can significantly aid future cost-effectiveness, healthcare policy, operations, and quality control studies.