EXPERIMENTS IN CONCEPT MODELING FOR RADIOGRAPHIC IMAGE REPORTS

Objective: Development of methods for building concept models to suppo rt structured data entry and image retrieval in chest radiography. Des ign: An organizing model for chest-radiographic reporting was built by analyzing manually a set of natural-language chest-radiograph reports . During model building, clinician-informaticians judged alternative c onceptual structures according to four criteria: content of clinically relevant detail, provision for semantic constraints, provision for ca nonical forms, and simplicity. The organizing model was applied in rep resenting three sample reports in their entirety. To explore the poten tial for automatic model discovery, the representation of one sample r eport was compared with the noun phrases derived from the same report by the CLARIT natural-language processing system. Results: The organiz ing model for chest-radiographic reporting consists of 62 concept type s and 17 relations, arranged in an inheritance network. The broadest t ypes in the model include FINDING, ANATOMIC LOCUS, PROCEDURE, ATTRIBUT E, and STATUS. Diagnoses are modeled as a subtype of FINDING. Represen ting three sample reports in their entirety added 79 narrower concept types. Some CLARIT noun phrases suggested valid associations among sub types of FINDING, STATUS, and ANATOMIC LOCUS. Conclusions: A manual mo deling process utilizing explicitly stated criteria for making modelin g decisions produced an organizing model that showed consistency in ea rly testing. A combination of top-down and bottom-up modeling was requ ired. Natural-language processing may inform model building, but algor ithms that would replace manual modeling were not discovered. Further progress in modeling will require methods for objective model evaluati on and tools for formalizing the model-building process.