Prediction rule for characterization of hepatic lesions revealed on MR imaging: estimation of malignancy.

OBJECTIVE Our aims were to establish factors that are most predictive of hepatic lesion malignancy and to formulate a prediction rule. MATERIALS AND METHODS A cross-sectional study of 227 abdominal MR imaging examinations revealed 85 lesions in 67 patients (29 men, 38 women; age range, 29-78 years; mean age, 51.4 years) who were being examined for primary malignancy (n = 42) or unknown lesion characterization (n = 25). All were referred for MR imaging after CT or sonography. Patient demographics (age, sex, history of malignancy), lesion size and morphology, quantitative T2 calculation, and pattern of enhancement on gadopentetate dimeglumine administration were evaluated for predictive ability. RESULTS Thirty-two liver lesions were malignant (eight colon cancer, five breast cancer, four cervical cancer, three renal cancer, three lung cancer, and nine miscellaneous cancers), 53 were benign (37 hemangiomas, 15 cysts, and one focal nodular hyperplasia). Calculated T2 relaxation times (mean +/- standard deviation [SD]) were as follows: malignant tumors (91.72 +/- 21.9 msec), hemangiomas (136.1 +/- 26.3 msec), cysts (284.1 +/- 38.2 msec) (p < 0.001). Logistic regression analysis indicated that lesion size and sex and age of patient were not significant independent predictors (p > 0.05). However, the combination of a history of malignancy, T2 value, and gadopentetate dimeglumine-enhancement pattern allowed generation of a prediction rule with an area under the receiver operating characteristic curve of 0.95. The patient's weight, lesion morphology, and cell type of the primary malignancy did not provide additional predictive information (p > 0.2). CONCLUSION We recommend using the combination of T2 quantification and patient history of malignancy before deciding to administer gadopentetate dimeglumine for optimal lesion characterization, especially for equivocal lesions with T2 values between 90 and 130 msec. These factors allowed the construction of a prediction rule for lesion characterization.