Extracting information from free text radiology reports

Abstract

We describe a system designed to extract and structure the information contained in free text radiology reports. The system uses natural language processing techniques to transform free text descriptions into structured information units that can be used to index and provide access to image databases. To provide access to the results of the system we developed a unique user interface for reporting and accessing radiology reports. We present a general system architecture for information extraction. Although the prototype focuses on thoracic radiology, this system can be extended into other areas of medicine. We discuss the system's knowledge sources and structures, and explain how natural language processing techniques are used to extract information from reports. The results of clinical testing of the prototype are presented and evaluated.