Differentiation of malignant and benign adrenal lesions with delayed CT: Multivariate analysis and predictive models

OBJECTIVE. The purpose of this study is to identify imaging and patient parameters that affect the diagnostic performance of delayed contrast-enhanced CT for distinguishing malignant from benign adrenal lesions larger than 1 cm in adult patients and to derive predictive models. MATERIALS AND METHODS. This retrospective study assessed 97 pathologically proven adrenal lesions that had undergone unenhanced, portal venous, and 15-minute delayed CT. Quantitatively, single-parameter evaluations of lesion attenuation (in Hounsfeld units) and absolute percentage enhancement washout (APEW) and relative percentage enhancement washout (RPEW) were performed. In addition, descriptive CT features (lesion size, margin defnition, heterogeneity vs homogeneity, fat, and calcifcation) and patients' demographic characteristics and medical history of malignancy were evaluated for association with lesion status using multiple logistic regression with stepwise model selection. Areas under the ROC curve (A z) were determined for univariate and multivariate analyses. Leave-one-lesion-out cross-validation was applied to ascertain the predictive performance of single-parameter and multivariate evaluations. RESULTS. The A z values for unenhanced attenuation, portal venous attenuation, delayed attenuation, APEW, and RPEW were 0.835, 0.534, 0.847, 0.792, and 0.871, respectively. Multivariate analyses revealed that portal venous attenuation, delayed attenuation, and APEW were signifcant features, with an Az of 0.923 when combined. The addition of the descriptive CT features increased the A z to 0.938; patient age and a history of malignancy were additional signifcant factors, increasing the Az to 0.956 and 0.972, respectively. The combined predictive classifer yielded 89% accuracy under cross-validation, compared with the best commonly applied single-parameter evaluation (77% for RPEW < 40%). CONCLUSION. Multivariate imaging evaluation applied to delayed contrast-enhanced CT alone, with or without patient characteristics, improves diagnostic performance for characterizing adrenal lesions beyond those of single-parameter evaluations. Predictive formulas assessing the probabilities of lesion benignity or malignancy are provided.