Progression-Free Survival Prediction for Locally Advanced Cervical Cancer After Chemoradiotherapy With MRI-based Radiomics.

Abstract

Aims: A significant proportion of locally advanced cervical cancer (LACC) patients experience disease progression post chemoradiotherapy (CRT). Currently existing clinical variables are suboptimal predictors of treatment response. This study reported a radiomics-based model leveraging information extracted from magnetic resonance (MR) T2-weighted image (T2WI) to predict the progression-free survival (PFS) for LACC following CRT.

Materials and methods: Radiomics features were extracted from pre-treatment MR T2WI in 105 LACC patients. Following pre-feature selection and a step forward feature selection method, an optimal feature set was determined with a Cox proportional hazard (CPH) model. The PFS predictions were generated through a radiomics-clinical combined model utilized five repeated nested 5-fold cross-validation (5-fold CV). Disease progression risk was stratified into high- and low-risk groups based on the predicted PFS and assessed by Kaplan-Meier analysis.

Results: The radiomics texture feature extracted from MR T2WI significantly predict PFS in LACC after CRT. In comparison with the model using clinical variables alone, the radiomics-clinical combined model achieves significantly improved performance in testing patient cohort, achieving higher C-index (0.748 vs 0.655) and area under the curve (0.798 vs 0.660 for 2-year PFS). Meanwhile, the proposed method significantly differentiated the high- and low-risk patients groups for disease progression (P < 0.001).

Conclusion: An MR T2WI-based radiomics and clinical combined model provided improved prognostic capabilities in predicting the PFS for LACC patients treated with CRT, outperforming a model using clinical variables alone. The incorporation of MR T2WI-based radiomics is promising in assisting in personalized management in LACC, indicating the potential of MR T2WI radiomics as imaging biomarker.