Preoperative prediction of parametrial invasion in early-stage cervical cancer with MRI-based radiomics nomogram

To develop and identify a MRI-based radiomics nomogram for the preoperative prediction of parametrial invasion (PMI) in patients with early-stage cervical cancer (ECC). All 137 patients with ECC (FIGO stages IB–IIA) underwent T2WI and DWI scans before radical hysterectomy surgery. The radiomics signatures were calculated with the radiomics features which were extracted from T2WI and DWI and selected by the least absolute shrinkage and selection operation regression. The support vector machine (SVM) models were built using radiomics signatures derived from T2WI and joint T2WI and DWI respectively to evaluate the performance of radiomics signatures for distinguishing patients with PMI. A radiomics nomogram was drawn based on the radiomics signatures with a better performance, patient's age, and pathological grade; its discrimination and calibration performances were estimated. For T2WI and joint T2WI and DWI, the radiomics signatures yielded an AUC of 0.797 (95% CI, 0.682–0.911) vs 0.946 (95% CI, 0.899–0.994), and 0.780 (95% CI, 0.641–0.920) vs 0.921 (95% CI, 0.832–1) respectively in the primary and validation cohorts. The radiomics nomogram, integrating the radiomics signatures from joint T2WI and DWI, patient's age, and pathological grade, showed excellent discrimination, with C-index values of 0.969 (95% CI, 0.933–1) and 0.941 (95% CI, 0.868–1) in the primary and validation cohorts, respectively. The calibration curve showed a good agreement. The radiomics nomogram performed well for the preoperative prediction of PMI in patients with ECC and may be used as a supplementary tool to provide individualized treatment plans for patients with ECC.