The Value Of Comparing Radiomics With Traditional Image Features To Predict Histological Grading Of Epithelial Ovarian Cancer

Objective: To compare the diagnostic performance of clinical model,conventional MRI morphological model,ADC model,T2 WI,DWI,DCE-MRI imaging histological model,traditional model,multi sequence imaging histological model and hybrid model in predicting type Ⅰ and type Ⅱepithelial ovarian cancer(EOC),and to verify whether the combination of traditional model and imaging histological model is more helpful in predicting the histological grade of EOC.Methods: From January 2015 to June 2019,61 patients with EOC confirmed by postoperative pathology in the Affiliated Hospital of Inner Mongolia Medical University were analyzed retrospectively,with 80 lesions(type Ⅰ = 30,type Ⅱ= 50)in total.All MRI examinations were performed on a 3.0T superconducting MR scanner(discovery MR750;GE Healthcare)using an 8-channel phase matrix coil.The main imaging sequences included axial fs-t2 wi,DWI and DCE-MRI.The clinical model,the conventional MRI morphological model,the ADC model and the traditional model combined with all parameters were established.The image group model including fs-t2 wi,DWI and T1 enhanced images was established,and the combined image group model was established by combining multiple sequences.1070 image group features are extracted from each sequence,and single variable analysis and least absolute contraction selection operator(lasso)are used to screen important features.A hybrid model is established by combining the traditional model with the multi sequence image group model.The difference of AUC value between different models was compared by.The difference of AUC value between different models was compared by Delong test.Accuracy,specificity,sensitivity,positive predictive value and negative predictive value were calculated according to the critical value of the maximum Yoden index.A calibration curve is used to evaluate the predictive performance of each model.The decision curve analysis(DCA)was carried out to evaluate the clinical applicability of each model by evaluating the net income under the threshold probability.Due to the small amount of data in this study,in order toensure that it is not affected by data deviation or insufficient sample size,all the above models are established by logistic regression without grouping,but by 10 fold cross validation and repeated modeling 20 times.Results:In the clinical model,age(P < 0.05)and CA125(P <0.05)represent independent predictors of typeⅡEOC.Based on the above independent variables,the AUC of the clinical model is 0.82(95% CI: 0.73-0.92).Tumor character(P <0.001)and ascites(P < 0.005)represent independent predictors of typeⅡEOC.AUC of conventional MR morphological model based on the above independent factors is 82%(95%CI: 0.73-0.92).The ADC value of type Ⅱtumor(9.3 ± 1.87 × 10-4 mm2 / s)was significantly lower than that of type Ⅰ tumor(12.34 ± 2.29 × 10-4 mm2 / s,P < 0.001).The AUC of ADC model was 86%(95% CI: 0.78-0.94).Among the logistic regression models of T2 WI,DWI and T1,DWI has a better predictive effect,AUC 81%(95% CI: 0.71-0.91).The results of multivariate logistic regression showed that age,CA125,tumor character,ascites and ADC were the most significant differences.Then,traditional logistic regression model was established to predict EOC histological grade.In ROC analysis,it was confirmed that the traditional model had a discrimination effect,AUC was 96%(95% CI: 0.92-1),and the combined multisequence image group model(AUC = 0.84,95% CI = 0.75-0.92)did not show a higher diagnostic effect.Delong test showed that the diagnostic efficiency of hybrid model(AUC = 0.97,95% CI = 0.93-1)was not significantly different from that of traditional model.The calibration curve shows that the traditional model and the hybrid model have similar prediction performance.DCA confirmed the clinical practicability of the two methods.The reliability and repeatability of the results are verified by cross validation.Conclusion:The diagnostic efficiency of the combined multisequence image group model is not higher than that of the traditional model,and there is no significant difference between the hybrid model and the traditional model,so the traditional model can be used as an effective tool to distinguish the type Ⅰ and type Ⅱclinical decision-making of EOC.