CT and MRI findings of type I and type II epithelial ovarian cancer.

OBJECTIVE

To assess whether types I and II epithelial ovarian cancer (EOC) differ in CT and MRI imaging features.

METHODS

For this retrospective study, we enrolled 65 patients with 68 ovarian lesions that have been pathologically proven to be EOC. Of these patients, 38 cases underwent MR examinations only, 15 cases underwent CT examinations only, and 12 cases completed both examinations. The clinical information [age, CA-125, menopausal status, and Ki-67] and imaging findings were compared between two types of EOCs. The diagnostic performance of image findings were assessed by receiver-operating characteristic curve(ROC) analysis. The association between EOC type and imaging features was assessed by multivariate logistic regression analysis. The random forest approach was used to build a classifier in differential diagnosis between two types of EOCs.

RESULTS

Of the 68 EOC lesions, 24 lesions were categorized as types I and other 44 lesions as type II based on the immunohistochemical results, respectively. Patients in type I EOCs were more likely to involve menopausal women and showed lower CA-125 and Ki-67 values (Ki-67<30%) than patients in type II EOCs. The imaging characteristics of type II EOCs frequently demonstrated a solid or predominantly solid mass (38.6% vs. 12.5%, P<0.05), smaller lesions (diameter <6cm; 27.3% vs. 4.2%, P<0.05), absence of mural nodules (65.9% vs. 25.9%, P=0.001), and mild enhancement (84.1% vs. 54.2%, P<0.05) compared to type I EOCs. Combination of tumor size, morphology, mural nodule, enhancement degrees (AUC=0.808) has a higher specificity (87.50%) and positive predictive value (90.0%) than any single image finding alone in differential diagnosis between two types of EOCs. The multivariate logistic regression analysis showed that enhancement degrees(OR 0.200, P<0.05),mural nodule(OR 0.158, P<0.05) significantly influence EOC classification. Random forests model identified both as the most important discriminating variables. The diagnostic accuracy of the classifier was 73.53%.

CONCLUSIONS

Differences in imaging characteristics existed between two types of EOCs. Combination of several image findings improved the preoperative diagnostic performance, which is helpful for the clinical treatment and prognosis evaluation.