Uncertainty/information as measure of various urographic parameters: an information theory model of diagnosis of renal masses.

Abstract

A new decision model for differentiating renal masses by urographic criteria has been developed and tested to demonstrate the use of uncertainty/information as a measure. The conditional probabilities used in this model are the relative frequency of occurrence of 15 specific uroradiographic signs in the presence of cyst, tumor, and benign cortical nodule. The decision model was applied to data obtained at the time of initial interpretation of 80 cases of renal mass discovered on urography. The probability of diagnosis was calculated by computer and compared to the radiologist's subjective probability estimate made prospectively at the time of initial interpretation. Information theory was applied to optimize the sequence in which signs were evaluated. The signs likely to maximally reduce the uncertainty of a diagnosis were evaluated first. The utility of this model and the comparative significance of various urographic signs used diagnose renal cysts, tumors, and benign cortical nodules were assessed. This model of renal mass evaluation at urography demonstrates principles of information theory that can be applied to more difficult and complex diagnostic and management problems.