DCE-MRI histogram analysis for predicting survival in patients with epithelial ovarian carcinoma

To evaluate the potential of dynamic contrast-enhanced MRI (DCE-MRI) histogram analysis for predicting overall survival (OS) in epithelial ovarian cancer (EOC) patients.

This retrospective study included 155 patients with EOC who underwent pretreatment DCE-MRI. Histogram features extracted from DCE-MRI parametric maps were selected using least absolute shrinkage and selection operator (LASSO) Cox regression, with selection stability assessed across 200 bootstrap replicates. Clinical, imaging, and combined models were developed using Cox regression and random survival forest (RSF). Apparent concordance indices (C-indices) were calculated and corrected for optimism using 100 bootstrap resamples with fixed predictor sets. Exploratory Kaplan-Meier analyses assessed risk-group separation, and SHapley Additive exPlanations (SHAP) analysis was used for model interpretation.

Median OS was 54 months. The 10th percentile of K^trans was selected in 88.0% of bootstrap replicates. Imaging Cox and RSF models achieved apparent C-indices of 0.790 and 0.827, respectively. Combined models incorporating this feature, International Federation of Gynecology and Obstetrics (FIGO) stage, and residual disease status achieved apparent C-indices of 0.806 for Cox regression and 0.853 for RSF, respectively. Bootstrap-corrected C-indices were higher for the combined than the corresponding clinical models, with a statistically significant improvement for Cox regression (0.805 vs. 0.737, P = 0.020) but not for RSF (0.810 vs. 0.745, P = 0.120). The combined models did not differ significantly (P = 0.820). Exploratory Kaplan-Meier analysis using cohort-specific median risk-score cutoffs showed significant OS separation (both P < 0.001). The 10th percentile of K^trans yielded the highest mean absolute SHAP value in the combined Cox model and exhibited the widest distribution of SHAP values in the combined RSF model.

DCE-MRI histogram features, particularly the 10th percentile of K^trans, may aid OS prediction in EOC and provide complementary prognostic information when integrated with clinicopathological factors. These exploratory findings require prospective multicenter external validation.