Artificial Neural Network In Preoperative Diagnosis Of Ovarian Mass

Background and ObjectiveOvarian masses are a common problem in gynecology,which can occur in all ages and vary in pathological histotype.However,there is still no effective screening and early diagnosis methods of Ovarian masses.Most patients with ovarian malignant tumor have been diagnosed with the disease at an advanced stage at the first visit to a doctor.The curative effect and prognosis of ovarian malignant tumor are poor and the mortality rate of it is highest among gynecological tumor.Since certainty about the nature of an adnexal mass can only be gained after histological examination,the traditional strategy for establishing the final diagnosis has been to perform an exploratory laparotomy.However,the preoperative diagnosis is based on doctor’s experience,which is largely subjective.The ability to provide accurate prognostic and predictive information to patients is becoming increasingly important as clinicians enter an era of personalized medicine.In order to avoid the subjectivity and mindset of clinicians in preoperative diagnosis,artificial neural networks have been widely used in the construction of predictive models for various diseases in recent years.Currently,there is no domestic literature report on preoperative diagnosis model of ovarian mass based on artificial neural network.The aim of this study was to generate and evaluate artificial neural network(ANN)models and logistics regression(LR)models from simple clinical and ultrasound-derived criteria to predict whether or not an ovarian mass will have histological evidence of malignancy.Methods1.The data were collected retrospectively from 382 patients with ovarian masses who have undergone surgical investigations at the Shandong Provincial Hospital,between January 2016 and September 2018.Age,parity history,symptoms,menopausal status,serum tumor marker levels and sonographic features of the adnexal mass were encoded as variables.The Descriptive and univariate analysis of variables are done by spss 25.2.Generate the logistics regression model(Logistic1)and BP-ANN model(BP-ANN1)for diagnosing ovarian masses from the serum tumor markers,including CA125,CA199,HE4 and ROMA.Create the logistics regression model(Logistic2)and BP-ANN model(BP-ANN2)for diagnosing ovarian masses from age,parity history,symptoms,menopausal status,serum tumor marker levels and sonographic features of the adnexal mass.All models are scripted by python language and run under Python 3.7.3.3.The performance of each model was evaluated using receiver operating characteristic(ROC)curves and area under curve(AUC).Result1.The AUC of the BP-ANN2 model is larger than that of the BP-ANN1 model;The AUC of the Logistic2 model is larger than that of the Logistic1 model.2.The AUC of the BP-ANN1 model is larger than that of the Logistic1 model.The AUC of the BP-ANN2 model is larger than that of the Logistic2 model.The AUC of BP-ANN2 for diagnosing ovarian benign masses,borderline tumors and ovarian malignancies are all larger than 0.9.Conclusion1.ANN combining with age,parity history,symptoms,menopausal status,serum tumor marker levels and sonographic features of the adnexal mass to build models and then to diagnose ovarian masses is better than tumor markers combining detection.The ANN models can well differentiate the ovarian benign masses,borderline tumors and ovarian malignancies.2.The accuracy of ANN diagnostic models for ovarian masses is higher than that of the logistics regression models.ANN is more suitable for analysis based on clinical data.