AUTOMATIC SYNTACTIC PARSING OF RADIOLOGICAL DIAGNOSES

We constructed a syntactic parsing system as an intelligent reporting interface for radiological diagnosis of the alimentary tract, into which radiological diagnoses could be put using ordinary language after which they were recognized and categorized automatically. The input diagnoses were limited to disease names, including both pathological and anatomical entities, and purely radiological description was excluded. The system was equipped with a table of possible diagnostic expressions composed of hundred of terms used in routine radiological work. A total of 95% of the CT and MRI diagnoses actually input by three radiologists were correctly recognized and corresponded to the code of the modified ''Index for Radiological Diagnoses of the American College of Radiological Diagnoses of the American College of Radiology.'' The remaining 5% were not categorized because of either over- or underdescription. This method was quicker and more user-friendly than hierarchical menu selection by mouse and/or Windows.