Bayesian Classifier for Predicting Malignant Renal Cysts on MDCT: Early Clinical Experience

Abstract

OBJECTIVE. The objective of our study was to evaluate the feasibility and usefulness of the Bayesian classifier for predicting malignant renal cysts on MDCT. MATERIALS AND METHODS. Ninety-three complicated cysts with pathologic confirmation were enrolled. Patient age and sex and seven morphologic features of the cysts including the maximum diameter, wall features, wall thickness, septa features, measurable enhancement of the wall and septa, presence of calcification, and presence of an enhancing soft-tissue component were used to train the Bayesian classifier. Four radiologists independently reviewed the MDCT images, and the probability of malignancy in each cyst was rated by the radiologists and the Bayesian classifier. The diagnostic performances of the radiologists visual decisions and the Bayesian classifier were then compared using receiver operating characteristic (ROC) curve analysis. The sensitivity and specificity were also compared between the visual decisions and the Bayesian classifier. RESULTS. The area under the ROC curve for predicting malignant renal cysts by the Bayesian classifier was greater than the visual decisions of three readers (reader 1, p = 0.02; reader 2, p < 0.01; reader 4, p = 0.02) and was similar to the visual decision of one reader (reader 3, p = 0.51). The specificity for predicting malignant renal cysts was greater by the Bayesian classifier than by the visual decisions in readers 2 (p = 0.04) and 4 (p = 0.02) and was similar in readers 1 (p = 0.68) and 3 (p = 1.00). In terms of sensitivity, there was no significant difference between the Bayesian classifier and the visual decisions in all four readers (p > 0.05). CONCLUSION. For predicting malignant renal cysts on MDCT, the Bayesian classifier is feasible and may improve diagnostic performance.