Multiparametric magnetic resonance imaging-based radiomics nomogram for predicting tumor grade in endometrial cancer

Abstract

Background: Tumor grade is associated with the treatment and prognosis of endometrial cancer (EC). The accurate preoperative prediction of the tumor grade is essential for EC risk stratification. Herein, we aimed to assess the performance of a multiparametric magnetic resonance imaging (MRI)-based radiomics nomogram for predicting high-grade EC.

Methods: One hundred and forty-three patients with EC who had undergone preoperative pelvic MRI were retrospectively enrolled and divided into a training set (n =100) and a validation set (n =43). Radiomic features were extracted based on T2-weighted, diffusion-weighted, and dynamic contrast-enhanced T1-weighted images. The minimum absolute contraction selection operator (LASSO) was implemented to obtain optimal radiomics features and build the rad-score. Multivariate logistic regression analysis was used to determine the clinical MRI features and build a clinical model. We developed a radiomics nomogram by combining important clinical MRI features and rad-score. A receiver operating characteristic (ROC) curve was used to evaluate the performance of the three models. The clinical net benefit of the nomogram was assessed using decision curve analysis (DCA), net reclassification index (NRI), and integrated discrimination index (IDI).

Results: In total, 35/143 patients had high-grade EC and 108 had low-grade EC. The areas under the ROC curves of the clinical model, rad-score, and radiomics nomogram were 0.837 (95% confidence interval [CI]: 0.754-0.920), 0.875 (95% CI: 0.797-0.952), and 0.923 (95% CI: 0.869-0.977) for the training set; 0.857 (95% CI: 0.741-0.973), 0.785 (95% CI: 0.592-0.979), and 0.914 (95% CI: 0.827-0.996) for the validation set, respectively. The radiomics nomogram showed a good net benefit according to the DCA. NRIs were 0.637 (0.214-1.061) and 0.657 (0.079-1.394), and IDIs were 0.115 (0.077-0.306) and 0.053 (0.027-0.357) in the training set and validation set, respectively.

Conclusion: The radiomics nomogram based on multiparametric MRI can predict the tumor grade of EC before surgery and yield a higher performance than that of dilation and curettage.

Keywords: endometrial cancer; histological grade; magnetic resonance imaging; nomogram; radiomics.

Figure 1

Flowchart of included and excluded patients with endometrial cancer.

Figure 2

Workflow of radiomics analysis and model building.

Figure 3

Eleven robust radiomics features and corresponding coefficients for rad-score construction.

Figure 4

Nomogram for predicting the tumor grade of endometrial cancer, established based on multiparameter magnetic resonance imaging and patient age.

Figure 5

Receiver operating characteristic curves of the three models predicting high-grade endometrial cancer in the training (A) and validation sets (B). The graphs (C) and (D) show that the calibration curve of nomogram has good calibration ability in both the training and validation sets, respectively.

Figure 6

(A) The clinical decision curve demonstrated that nomogram has higher net benefits than preoperative curettage at a threshold probability of 0–0.47 and > 0.67. The solid blue and orange lines in figures (B) and (C) represent the clinical impact curves of the nomogram and the actual preoperative DC, respectively. The black dashed line represents the postoperative pathological results of patients with endometrial cancer, and the closer the solid line is to the black dashed line, the better the prediction effect.