The Clinical Value Of Ultrasound Models In Evaluating The Risk Of Malignant Ovarian Tumors

ObjectivesTo compare the clinical value of the International Ovarian Tumor Analysis(IOTA)Simple rules(SR),Assessment of Different NEoplasias in the adne Xa(ADNEX)model,subjective assessment by experts(SA)and artificial intelligence(AI)in evaluating the risk of malignant ovarian tumors.MethodsA total of 278 patients admitted for surgery for ovarian(including parovarian and fallopian tube)tumors between May and December 2017 were included.Two gynecological and obstetric ultrasound experts with more than 10 years of experience assessed the sonographic tumor morphology based on the terms and measurement methods of the IOTA ultrasound diagnostic consensus,and the clinical value of SR,ADNEX model,SA and AI for differentiation between benign and malignant ovarian tumors were compared.All patients underwent surgery and the histopathological diagnosis was used as reference standard.The sensitivity,specificity,positive predictive value,negative predictive value,positive likelihood ratio,negative likelihood ratio,diagnostic odds ratio,and area under the ROC curve were used to evaluate the diagnostic performance.ResultsOf the 278 women included,203(73.02%)had benign ovarian tumors and 75(26.98%)had malignant ovarian tumors,including 18(6.47%)with BOT,17(6.12%)with Stage-I OC,32(11.51%)with Stages-II-IV OC and eight(2.88%)with ovarian metastasis.In all cases,SR can be used for benign and malignant classification accounted for 86.69%.If the inconclusive cases were classified as malignant,the sensitivity and specificity of SR were 89.33% and 86.21%.The performance of the IOTA ADNEX model was good for discriminating between benign and malignant tumors,with an area under the ROC curve(AUC)of 0.94(95%CI,0.91-0.97)when CA125 was included in the model and AUC of 0.93(95%CI,0.90-0.96)without CA125,and there was no statistical difference between the two(P=0.54).When10% was used as the prediction of malignant risk,the sensitivity was 93.33%,and the specificity was 77.83%.The AUC values of the model including CA125 ranged between0.61 and 0.99 for distinguishing between the different types of tumor,and it showed excellent performance in discriminating between benign ovarian tumors and Stages-II-IV OC,with an AUC of 0.99(95%CI,0.97-1.00).The performance of the model was less effective at distinguishing between BOT and Stage-I OC and between Stages-II-IV OC and ovarian metastasis,with AUC values of 0.61(95%CI,0.43-0.77)and 0.78(95%CI,0.62-0.90).The sensitivity and specificity of SA in differential diagnosis of benign and malignant ovarian tumors were 96.00% and 91.63%,and the AUC was 0.97(95% CI,0.94-0.98).The sensitivity and specificity of AI were 91.89% and 91.67%,and the AUC was 0.98(95% CI,0.94-0.99).The diagnostic efficacy of AI was not statistically different from that of SA(P=0.362),and was better than that of ADNEX model(P=0.049)and SR.ConclusionsAmong the various methods for evaluating the risk of malignant ovarian tumors,AI had excellent diagnostic efficiency,with no statistical difference compared with SA,and was better than IOTA SR and ADNEX model.IOTA SR was simple and practical,but only 86.69% of ovarian tumors could be classified by simple rules.ADNEX model not only showed high diagnostic efficiency in differentiating benign and malignant ovarian tumors,but also capable of subclassifying the tumors,however,the discriminative performances of some malignant subtypes were less effective.