Multiparametric MRI-based Deep Learning and Radiomics for Evaluating Lymph Node Metastasis in Early-Stage Cervical...

Purpose To develop a multiparametric MRI-based radiomics model and deep learning-radiomics (DLR) fusion model for preoperative prediction of lymph node metastasis (LNM) in early-stage cervical cancer. Materials and Methods In this multicenter retrospective study (January 2020-December 2022), preoperative MRI data from patients with early-stage cervical cancer were split into training, internal testing, and external testing cohorts. Radiomic and deep learning (DL) features of both the tumor and lymph node were extracted separately from the MRI scans. Multivariable logistic regression was used to construct predictive models for LNM based on tumor and lymph node radiomic features (Rad_T+LN) and based on radiomic and DL features from both the tumor and lymph node (DLR_T+LN). The models' effectiveness and clinical applicability were evaluated using receiver operating characteristic curves, calibration curves, and decision curve analysis. A two-tailed P value of <.05 was considered statistically significant. Results The overall dataset included 862 patients (median age, 53 years [IQR, 45-60 years]). Rad_T+LN resulted in areas under the receiver operating characteristic curve (AUCs) of 0.81 (95% CI: 0.76, 0.86), 0.79 (95% CI: 0.72, 0.87), and 0.77 (95% CI: 0.71, 0.82) in the training, internal testing, and external testing cohorts, respectively. DLR_T+LN achieved AUCs of 0.83 (95% CI: 0.76, 0.91) and 0.79 (95% CI: 0.74, 0.84) in the internal and external testing cohorts, respectively, and did not improve over Rad_T+LN (P > .05). Both models demonstrated good calibration and positive net benefit on decision curve analysis. Conclusion Rad_T+LN and DLR_T+LN exhibited robust diagnostic performance for LNM prediction. Keywords: MR-Diffusion Weighted Imaging, MR Imaging, Genital/Reproductive, Cervix, Metastases, Decision Analysis, Segmentation, Radiomics, Diagnosis, Uterine Cervical Neoplasms, Lymphatic Metastasis, Magnetic Resonance Imaging, Deep Learning Supplemental material is available for this article. © RSNA, 2026.