Radiomics Analysis of Multiparametric MRI for the Preoperative Prediction of Lymph Node Metastasis in Cervical Cancer

Abstract

Objective: To develop and validate a radiomics predictive model based on multiparameter MR imaging features and clinical features to predict lymph node metastasis (LNM) in patients with cervical cancer. Material and Methods: A total of 168 consecutive patients with cervical cancer from two centers were enrolled in our retrospective study. A total of 3,930 imaging features were extracted from T2-weighted (T2W), ADC, and contrast-enhanced T1-weighted (cT1W) images for each patient. Four-step procedures, mainly minimum redundancy maximum relevance (MRMR) and least absolute shrinkage and selection operator (LASSO) regression, were applied for feature selection and radiomics signature building in the training set from center I (n = 115). Combining clinical risk factors, a radiomics nomogram was then constructed. The models were then validated in the external validation set comprising 53 patients from center II. The predictive performance was determined by its calibration, discrimination, and clinical usefulness. Results: The radiomics signature derived from the combination of T2W, ADC, and cT1W images, composed of six LN-status-related features, was significantly associated with LNM and showed better predictive performance than signatures derived from either of them alone in both sets. Encouragingly, the radiomics signature also showed good discrimination in the MRI-reported LN-negative subgroup, with AUC of 0.825 (95% CI: 0.732-0.919). The radiomics nomogram that incorporated radiomics signature and MRI-reported LN status also showed good calibration and discrimination in both sets, with AUCs of 0.865 (95% CI: 0.794-0.936) and 0.861 (95% CI: 0.733-0.990), respectively. Decision curve analysis confirmed its clinical usefulness. Conclusion: The proposed MRI-based radiomics nomogram has good performance for predicting LN metastasis in cervical cancer and may be useful for improving clinical decision making.

Keywords: lymph nodes; magnetic resonance imaging; nomograms; predictive value of tests; uterine cervical neoplasms.

Copyright © 2020 Hou, Zhou, Ren, Du, Xin, Zhao, Cui and Zhang.

Figures

Figure 1

The workflow of radiomic analysis in the current study.

Figure 2

Plots (A–F) present the boxplots of the six radiomics feature with significant difference between the LN metastasis (LN+) and LN negative (LN–) groups in the training datasets, respectively.

Figure 3

Plots (A,B) show the Rad-score for each patient, plots (C,D) show the receiver operating characteristic (ROC) curves of the radiomics signature derived from T2W, ADC, and cT1W images and their combination, and plots (E,F) present the boxplots of the Rad-score in the training and validation sets, respectively.

Figure 4

The predictive performance of the radiomics signature in the MRI-reported LN-negative subgroup. Plots (A,B) show the ROC curve of the radiomics signature and the Rad-score of individual patients in the MRI-reported LN-negative subgroup.

Figure 5

Radiomics nomogram developed with ROC curves and calibration curves. (A) A radiomics nomogram was developed for the prediction of LNM in the training set, with radiomics signature and MRI-reported LN status incorporated. Comparison of ROC curves between the radiomics nomogram and MRI-reported LN status alone for the prediction of LN metastasis in the (B) training and (C) validation sets. Plots (D,E) present the calibration curves of the radiomics nomogram in the training and validation sets, respectively.

Figure 6

Decision curve analysis (DCA) for the radiomics nomogram and the MRI-reported LN status in the validation set. The y-axis represents the net benefit. The x-axis represents the threshold probability. The decision curves showed that if the threshold probability is over 10%, the application of radiomics nomogram to predict LNM adds more benefit than treating all or none of the patients and MRI-reported LN status.