Related Experiment Videos
Automatic extraction of PIOPED interpretations from ventilation/perfusion lung scan reports
M Fiszman1, P J Haug, P R Frederick
1Department of Medical Informatics, LDS Hospital, University of Utah, USA.
Proceedings. AMIA Symposium
|February 3, 1999
Summary
Natural language processing tools can effectively extract clinical information from Ventilation/Perfusion lung scan reports. This facilitates automated quality assurance and improves diagnostic performance monitoring in radiology departments.
Area of Science:
- Radiology Informatics
- Natural Language Processing
- Clinical Data Extraction
Background:
- Free-text radiology reports are difficult for computer systems to analyze.
- Existing natural language processing (NLP) methods have primarily focused on chest x-ray reports.
- Lack of structured data hinders medical decision support, quality assurance, and outcome studies.
Purpose of the Study:
- To evaluate the effectiveness of the SymText NLP tool for extracting clinical information from Ventilation/Perfusion (V/P) lung scan reports.
- To assess the performance of NLP in V/P reports compared to chest x-ray reports.
- To explore the potential for automated quality monitoring in radiology.
Main Methods:
- Utilized the SymText NLP tool to process V/P lung scan reports.
- Analyzed the extracted information for precision and recall.
- Compared NLP performance on reports with and without an impression section.
Main Results:
- Achieved high performance metrics: overall precision of 0.88 and recall of 0.92.
- Observed differences in NLP system function based on the presence of an impression section in reports.
- Demonstrated successful extraction of key clinical information from V/P reports.
Conclusions:
- NLP tools like SymText can accurately extract valuable clinical data from V/P lung scan reports.
- Automated data extraction enables the development of quality monitors for radiologist performance.
- Reduces manual effort for quality assurance, allowing for more frequent reviews.