CT of cystic renal masses: Analysis of diagnostic performance and interobserver variation

OBJECTIVE. Our objective was to assess the clinical usefulness and interobserver variability of the Bosniak classification scheme for characterizing a series of pathologically proven cystic renal lesions imaged with CT. MATERIALS AND METHODS. Seventy pathologically proven cystic renal masses (38 benign, 32 malignant) in 46 patients were reviewed independently by three radiologists. The cystic masses were categorized by each reviewer according to both the Bosniak classification and the receiver operating characteristic (ROC) analysis. Both the individual results for each reader and the pooled results for all three readers were analyzed. Interobserver agreement and discordance in classifying lesions as Bosniak categories I-II or III-IV were assessed. RESULTS. The distribution of the 70 lesions (based on the average of the three readers) was 22 Bosniak I (0% malignant), eight Bosniak II (13% malignant); 11 Bosniak III (45% malignant), and 29 Bosniak IV (90% malignant). All readers agreed on the Bosniak classification in 59%, or 41 of the 70 lesions (I, 17; II, one; III, four; and IV, 19). Eleven (16%) of the 70 lesions were classified as Bosniak I or II by one reader and as Bosniak III or IV by at least one other reader. The area under the curve for the pooled ROC analysis was calculated to be 0.957. Individual reader values ranged from 0.914 to 0.981. The sensitivities, specificities, and accuracies for the three readers ranged from 94% to 100%, 71% to 92%, and 84% to 93%. Assessment of interobserver variability by kappa analysis yielded scores of .571 and .477 for the Bosniak and ROC analyses, respectively. CONCLUSION. Overall, the Bosniak classification scheme is useful for evaluating renal masses; however, interobserver variation in distinguishing Bosniak II and Bosniak III lesions may present difficulties in recommending surgical versus conservative management.