Different multiparametric MRI-based radiomics models for differentiating stage IA endometrial cancer from benign endometrial lesions: A multicenter study

• DOI: 10.3389/fonc.2022.939930
• Publication Date: 2022-08-05T10:34:55Z
• Authors: Qiu Bi; Yaoxin Wang; Yuchen Deng; Yang Liu; Yuanrui Pan; Yang Song; Yunzhu Wu; Kunhua Wu
• Keywords: Nomogram; Univariate; Univariate analysis

The aim of this study was to evaluate the value different multiparametric MRI-based radiomics models in differentiating stage IA endometrial cancer (EC) from benign lesions.The data patients with lesions two centers were collected. features extracted T2-weighted imaging (T2WI), diffusion-weighted (DWI), apparent diffusion coefficient (ADC) map, and late contrast-enhanced T1-weighted (LCE-T1WI). After dimension reduction feature selection, nine machine learning algorithms conducted determine which optimal model for differential diagnosis. univariate analyses logistic regression (LR) performed reduce valueless clinical parameters develop model. A nomogram using radscores combined developed. Two integrated obtained respectively by ensemble strategy stacking algorithm based on area under curve (AUC), decisive (CDC), net reclassification index (NRI), discrimination (IDI) used performance benefits models.A total 371 incorporated. LR highest average AUC (0.854) accuracy (0.802) internal external validation groups (AUC = 0.910 0.798, respectively), outperformed 0.739 0.592, respectively) or radiologist 0.768 0.628, respectively). 0.917 0.802, achieved better than groups. 0.915) 0.918) had a similar compared group, whereas AUCs 0.792) 0.794) lower those group. According CDC, NRI, IDI, model, nomogram, good benefits.Multiparametric can non-invasively differentiate EC lesions, is best algorithm. presents excellent stable diagnostic efficiency.