Logistic regression analysis of combining transvaginal conventional ultrasonography and 3D power Doppler ultrasonography (3D-PDUS) in differential diagnosis of ovarian tumor

Abstract

Objective To develop a logistic regression model for differential diagnosis of the malignant ovarian tumor by combining transvaginal conventional ultrasonography and 3D power Doppler ultrasound(3D-PDUS) techniques. Methods The transvaginal ultrasonography and 3D-PDUS data were collected from 291 patients with ovarian tumors received clinical pathological diagnosis. According to the pathological resuts, the 291 patients were divided in to benign group(GB) and borderline/malignant group(GM). Univariate and multivariate logistic regression analyses were applied to establish models for predicting malignant tumor. Results ①Within the 291 ovarian tumor patients, 175 (60.14%) were classified as the GB and 116 (39.86%) were considered as the GM, the number for postmenopausal cases and the serum CA125 level in GB group were significantly lower than those in GM group (P<0.001); ②The tansvaginal conventional ultrasound analysis suggested that borderline/malignant tumor often presented as larger volume, more irregular shape, and higher incidence in the blood flow within a solid papillary projection and ascites (P<0.05, compared with GB group); In 3D-PDSU data, the vascularization index (VI) was lower than that in GB group (P<0.001, compared with GM group), but there was no significant difference between GB group and GM group in flow index (FI) and vascularization-flow index (VFI) (P=0.559, 0.454); ③Multivariate logistic regression analysis showed that postmenopausal status, serum CA125 levels, tumor echo, papillary with blood flow, ascites and 3D-PDUS VI parameters were independent risk factors for the progression of borderline/malignant tumor. The ROC curve showed that the established regression model accuracy was 92.0%, sensitivity was 86.2%, specificity was 95.7%, positive predictive value was 92.6% and negative predictive value was 91.7%. Conclusions Combining multi-model transvaginal conventional ultrasound and 3D-PDUS analyses is a useful non-invasive technique for the differential diagnosis of ovarian tumor. Key words: Endosonography; Three-dimensional power Doppler ultrasonography; Ovarian neoplasms; Logistic regression