Information retrieval for teaching files: a preliminary study

Abstract

A computer algorithm for information retrieval from an electronic teaching file has been developed. This index enables the user to retrieve cases from a teaching file, based on the input of a combination of features. The algorithm is based on nearest neighbor analysis, and is programmed in the "C" language. A teaching file with this index is very easy to use as a reference resource for diagnosing unknown cases. A model was developed for a preliminary test of how likely a user would be to review a teaching file case that is the same diagnosis as an unknown case, thereby reducing uncertainty of diagnosis. The model used 110 cases of arthritis radiographs of hands scored by a skeletal radiologist. The result of the model suggests that the correct diagnosis would be reviewed 83% of the time. A standard method of reducing uncertainty of diagnosis (the maximum likelihood discriminant function) would have picked the correct diagnosis 78% of the time. The results indicate that a teaching file with the computer index is a practical tool for dealing with the uncertainty in diagnosis of unknown cases. The computer index could be included with videodisc-based teaching files (such as the American College of Radiology files). Using teaching files as a reference for interpreting unknown cases may reduce interobserver variability.