Value of an MRI radiomics model in predicting early postoperative recurrence of esophageal cancer

YANG Rihui; ZHONG Yi; LIN Zhiping; ZHONG Fang; QIU Jinghua; ZHANG Tianhui; FAN Weixiong

doi:10.19300/j.2025.L21766

Abstract：Objective To explore the value of a predictive model based on MRI radiomics features for early postoperative recurrence in patients with esophageal cancer.Methods A total of 224 patients with esophageal cancer confirmed by postoperative pathological examination were retrospectively collected.All patients underwent preoperative multimodal MRI examination.Based on the 1-year postoperative follow-up,patients were divided into a recurrence group and a non-recurrence group.They were randomly assigned to a training set(157 cases;58 recurrence,99 non-recurrence)and a validation set(67 cases;14 recurrence,53 non-recurrence)in a 7∶3 ratio.Radiomics features were extracted from T2WI BLADE(motion-corrected)sequence and contrast-enhanced T1-weighted StarVIBE(free-breathing,radial K-space acquisition).Radiomics feature selection was performed,and Logistic regression was used to construct prediction models.The predictive performance of each model was evaluated using receiver operating characteristic(ROC)curve and the area under the curve(AUC)was calculate.Differences in AUC values were compared using the Delong test.Clinical net benefits of the model was assessed via decision curve analysis.Results Fourteen and 15 radiomics features were selected from the T2WI BLADE and T1WI StarVIBE sequences,respectively.Based on these features,T2WI BLADE radiomics models,T1WI StarVIBE enhanced radiomics models,and combined radiomics models were constructed.The AUC values of these three models in the training and validation sets were 0.778,0.889,0.920,and 0.699,0.884,0.911,respectively.Delong's test showed that the combined model had significantly better performance in predicting early recurrence than either single-sequence model(all P＜0.05).Decision curve analysis shows that the combined model provided greater clinical net benefit and closely aligned with clinical outcomes.Conclusion MRI-based radiomics features can effectively predict the risk of early postoperative recurrence in patients with surgically esophageal cancer.The combined radiomics model demonstrates superior predictive performance compared to single-sequence models.