The Study On Setting Up The Improved Risk Of Malignancy Model And Artificial Neural Network's Model In Ovarian Tumor

Objective To revaluate the performance of the three kinds of therisk of malignancy index (RMI) in diagnosing malignancy in women with known adnexal masses, and propose the reasons and directions of improving RMI's model.Methods One hundred and eighty women with known adnexal mass who had been examined before surgery were recruited from QiLu Hospital of Shan Dong University. The three kinds of RMI's models (The RMI is the product of menopausal score, ultrasound scores, and the absolute value of serum CA125 levels) that Professor Jacobs had founded and professor Tingulstad had modified two-timly, were prospectively applied to predict the likelihood of malignancy in 180 patients with adnexal masses. According to patients's histopathlogic results, their diagnostic values were analysed in the different histological types and FIGO stages of tumours, their diagnostic capability were compared with single indices by using ROC (Receiver Operating Characteristic) curve, their best suitable cut-off value was compared with result that original research reported.Results There were statistic significant differences in the diagnostic specificity between RMI's model (RMI196.09 %; RMI293.23%; RMI3 96.24%) and CA125 (77.44%), ultrasound scores(80.45%)(P<0.001),but there were not statistic significant differences in the diagnostic sensitivity between RMI's model (RMI165.96%;RMI270.21%; RMI3 68.08%) and CA125 (74.46%) , ultrasound scores(78.72%)(P>0.05); The diagnostic rates of three kinds of RMI's models in the nonepithelial malignant (50%) were significantly lower than in the epithelial malignant tumours (RMI183.33%; RMI291.67%; RMl387.50%)(P<0.05), and were particularly poorer in ovarian mucinous cystadenocarcinoma, borderline tumor, granular cell carcinoma and germ cell malignant tumor. There were many misdiagnosises in some benign tumors with high blood level of CA125. In the group of patients with borderline tumour and early stage tumor, the diagnostic capabilities of RMI were significantly lower than in the group of patients with later stage tumor (P<0.05). The area under the ROC curve of RMI (RMI10.815;RMI20.817;RMI30.822) were larger than that of menopausal score(0.691), CA125(0.756) and ultrasound score(0.800), but there were not statistic significant differences(P>0.05). There were no differences among the diagnostic capabilities of three kinds of models in the scale of 20-300 cut-off values (P=0.05), their best suitable cut-off value was level of 100 other than level of 200 that was originally reported.Conclusion The three kinds of RMI's model increase diagnosic specificity of discriminating benign and malignant adnexal masses, worthwhile they hold poor diagnosic sensitivity in ovarian nonepithelial tumor, borderline and early stage malignant tumor, and hold high misdiagnosis rate in some benign tumors that have high blood CA125 levels. So they need to be improved.Objective To develop a new risk of malignancy model which should have higher diagnostic efficiency by adding new parameters based on the risk of malignancy index (RMI).Methods One hundred and eighty women with known adnexal mass who had been examined before surgery were recruited from QiLu Hospital of ShanDong University.In addition to the parameters which had been studied in RMI's Model, TSGF (tumor supplied group of factors), adjusted ultrasound scores and the results of Doppler blood flow analysis were obtained before operation. Firstly the diagnostic values of every parameter were evaluted in all patients, and then the improved risk of malignancy model was worked out by using binary logistic regression. The performance of the improved risk of malignancy model in predicting malignancy of adnexal masses was compared with the risk of malignancy index by Receiver Operating Characteristic test. The diagnostic accuracy of the improved risk of model was evaluated in all patients with differents histological classification and FIGO stage adnexal malignant masses.Result All parameters including age,menopausal state, blood CA125 and TSGF levels, the adjusted ultrasound score , systolic peak velocity, end diastolic velocity, time-averaged mean velocity, pulsatility index and resistance index were significant in judging malignancy of adnexal masses (P=0.000).Theimproved risk of malignancy model (P =1/1+e-Z) was designed by using binarylogistic regression, z = (-15.025+0.073 ×age+0.083 ×TSGF+0.878 × the revised ultrasound+0.222 × time-averaged mean velocity). The area under the ROC curve of improved risk of malignancy model (0.885) was larger than three kinds of the risk of malignancy index (RMI10.815; RMI2 0.817; RMI30.822) (P<0.05), it achieved 80.85% sensitivity and 96.24% specificity. The diagnostic sensitivity of improved risk of malignancy model in ovarian malignant germ cell tumor and granular cell carcinoma (71.43%) was higher than of RMI (28.57%) (P=0.026); and in ovarian borderline tumors and early stage malignant tumors (66.67%), The diagnostic sensitivity of it was higher than that of RMI1 (33.33%).Conclusion The improved risk of malignancy model increases the diagnostic performance of RMI in predicting malignancy of adnexal masses. It not only holds higher diagnostic specificity in all adnexal malignant masses, but also raises diagnostic sensitivity in malignant germ cell tumor, granular cell carcinoma, borderline tumour and early stage malignant tumor. It should deserve to be validated and spreaded in clinical practice.Objective To set up a model for predicting malignancy in patients with adnexal masses by using computer intelligent technology- artificial neural network (ANN), and evaluate the diagnosic performance of ANN's model.Methods One hundred and eighty women with known adnexal masswho had been examined before surgery were recruited from QiLu Hospital of ShanDong University.The every patient's demographic, biochemical and sonographic data were recorded, those parameters included age; menopausal status; familial ovarian or lacteal gland malignant tumor history; other system malignant tumor history;gravidity; blood levels of CA125, CEA,SA, HCG, AFP, TSGF and FE; volume of tumor; ultrasound grey-scale morphological characteristics; and the results of Doppler blood flow analysis. Firstly, the significant single parameters were screened out from all patients, and then two more significant parameters which would act as imput variables of ANN's model were separately selected from each demographic date, biochemical date, sonographic ultrasound characteristic data and date of Doppler blood flow analysis by using binary logistic regression. According to the ratio of 7:3, the date of 180 patients with adnexal masses was randomly divided into training and testing subsets. The training subsets were used to set up ANN's model -MLP's model (Multi Layer Perceptron) based on the resultes of binary logistic regression. The testing subsets were used to estimate the performance among of the risk of malignancy index, the improved risk of malignancy model and ANN's model in predicting adnexal mass's malignancy by Receiver OperatingCharacteristic test. The diagnostic accuracy of ANN's model was evaluated in whole 47 patients with differents histological classification and FIGO stage adnexal malignant masses.Result Twenty of twenty-five parameters including age;menopausalstate; positive family history; blood levels of CA125, CEA, SA, HCG , AFP, TSGF and FE; site and boundary of tumor; echogenicity; wall thickness; presence of papillary projections and ascites; systolic peak velocity; end diastolic velocity; time-averaged mean velocity; pulsatility index and resistance index could help to judge malignant characteristic of adnexal masses (P<0.05). Three-layer MLP model, based on 8 input variables (included age, gravidity, papillary projection, ascites, systolic peak velocity and time-averaged mean velocity)selected by using binary logistic regression had a significant higher general diagnostic efficacy (the area under ROC curve was 0.940) than did RMI's model (RMI10.845; RMI20.857; RMI3 0.845) and the improved risk of malignancy model (0.893) (P<0.05), achieved 87.23% sensitivity and 97.74% specificity. In particular, it holded higher diagnostic rate than did RMI's model and the improved risk of malignancy model in the patients with ovarian nonepithelial malignant tumors (80.00%), and in borderline and early stage malignant tumors (88.89%) (P<0.05).Conclusion Computer intelligent technology--artificial neural network applied in predicting malignancy of adnexal masses has better diagnostic capability than does statistics methods. In particular, it holds higher diagnostic rate in the patients with ovarian nonepithelial malignant tumors, borderline tumors and early stage malignant tumors. There is a need for further consummate ANN's model.