Development And Validation Of An MRI-based Radiomics Nomogram For Assessing Deep Myometrial Invasion In Early Stage Endometrial Adenocarcinoma

Objective(s): To establish and verify a radiomics nomogram model for predicting deep myometrial invasion(DMI)in early-stage endometrioid adenocarcinoma(EAC)and its auxiliary value to radiologists.Methods: A total of 266 patients with stage I EAC were randomly divided into a 7:3 ratio to form the training(n=185)and test groups(n=81).T2-weighted imaging,diffusion-weighted imaging,apparent diffusion coefficient maps and contrast-enhanced T1-weighted imaging were performed by using 3.0T and 1.5T magnetic resonance scanners.Two radiologists outlined tumour contours as regions of interest and extracted 1781 radiomics features in each sequence separately.Intraclass Correlation Coefficient(ICC)was calculated within and between observers and features with ICC ≥ 0.75 and Pearson correlation coefficient < 0.9 were selected.The minimum absolute contraction and selection operator algorithms were used to select the appropriate radiomics features.Then establish the radiomics feature model and calculate the radiomics score(Radscore)according to the coefficients of each features.Univariate and multivariate logistic regression were used to identify independent clinical factors.The important clinical factors and radiomics features were integrated into a radiomics nomogram.A receiver operating characteristic curve was used to evaluate the radiomics nomogram.All patients were regrouped by MR scanner field strength for cross-validation.Two radiologists evaluated MR images with or without the help of the radiomics nomogram to detect the presence of DMI.The clinical benefit of using the radiomics nomogram was evaluated by decision curve analysis and by calculating the net reclassification improvement and integrated discrimination improvement.Results: 1.Age and CA125 were independent clinical predictors.2.(1)The area under the curves of the clinical parameters,radiomics signature and nomogram in evaluating DMI were 0.744,0.869 and 0.883,respectively.(2)The cross-validation results regrouped by MR scanner field strength showed that the AUC of the model constructed based on radiomics features was 0.850 and the AUC of the radiomics nomogram was 0.865,respectively.The Delong test showed that there was no significant difference between the original group and the regrouping group(p > 0.05).3.(1)The accuracies of the two radiologists increased from 79.0%and 80.2% to 90.1% and 92.5% when they used the nomogram.Kappa analysis showed good agreement between the two radiologists,both unassisted(kappa = 0.804,p < 0.001)and with the assistance of the nomogram(kappa = 0.882,p < 0.001).(2)The net reclassification improvement of the two radiologists were 0.262 and 0.318,and the integrated discrimination improvement were 0.322 and 0.405.(3)According to the decision curve,the nomogram showed a higher net benefit than the radiomics signature or unaided radiologists.Conclusion(s): The radiomics nomogram,based on radiomics features and clinical factors,can help radiologists evaluate DMI and improve their accuracy in predicting DMI in early-stage EACs and provide a basis for clinicians to develop individualised treatment plans preoperatively.The cross-validation test showed that changes in field strength did not significantly affect the machine learning results.