RESUMEN
OBJECTIVE: To conduct multiparametric magnetic resonance imaging (MRI)-derived radiomics based on multi-scale tumor region for predicting disease-free survival (DFS) in early-stage squamous cervical cancer (ESSCC). METHODS: A total of 191 ESSCC patients (training cohort, n = 135; validation cohort, n = 56) from March 2016 to September 2019 were retrospectively recruited. Radiomics features were derived from the T2-weighted imaging (T2WI), contrast-enhanced T1-weighted imaging (CET1WI), diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) map for each patient. DFS-related radiomics features were selected in 3 target tumor volumes (VOIentire, VOI+5 mm, and VOI-5 mm) to build 3 rad-scores using the least absolute shrinkage and selection operator (LASSO) Cox regression analysis. Logistic regression was applied to build combined model incorporating rad-scores with clinical risk factors and compared with clinical model alone. Kaplan-Meier analysis was used to further validate prognostic value of selected clinical and radiomics characteristics. RESULTS: Three radiomics scores all showed favorable performances in DFS prediction. Rad-score (VOI+5 mm) performed best with a C-index of 0.750 in the training set and 0.839 in the validation set. Combined model was constructed by incorporating age categorized by 55, Federation of Gynecology and Obstetrics (Figo) stage, and lymphovascular space invasion with rad-score (VOI+5 mm). Combined model performed better than clinical model in DFS prediction in both the training set (C-index 0.815 vs 0.709; p = 0.024) and the validation set (C-index 0.866 vs 0.719; p = 0.001). CONCLUSION: Multiparametric MRI-derived radiomics based on multi-scale tumor region can aid in the prediction of DFS for ESSCC patients, thereby facilitating clinical decision-making. KEY POINTS: ⢠Three radiomics scores based on multi-scale tumor region all showed favorable performances in DFS prediction. Rad-score (VOI+5 mm) performed best with favorable C-index values. ⢠Combined model incorporating multiparametric MRI-based radiomics with clinical risk factors performed significantly better in DFS prediction than the clinical model. ⢠Combined model presented as a nomogram can be easily used to predict survival, thereby facilitating clinical decision-making.