Preoperative evaluation of extramural venous invasion in rectal cancer using radiomics analysis of relaxation maps from synthetic MRI

To establish a radiomics nomogram based on relaxation maps for predicting the extramural venous invasion (EMVI) of rectal cancer (RC) and compare the diagnostic efficacy of the nomogram and subjective assessment by radiologists. Among 94 RC patients receiving direct surgical resection, 65 were randomly allocated to the training cohort and 29 to the validation cohort. Radiomics features were extracted from synthetic magnetic resonance imaging including T1, T2, and proton density (PD) maps. The least absolute shrinkage and selection operator methods were used for dimension reduction, feature selection, and radiomics model building. Multivariable logistic regression analysis was used for nomogram development. The performance of the nomogram was assessed with respect to its calibration, receiver operating characteristics (ROC) curve, and decision curve analysis. The radiomics model demonstrated good predictive efficacy for EMVI, with an area under the ROC curve (AUC), sensitivity, and specificity of 0.912 (95% confidence interval (CI), 0.837–0.986), 0.824, and 0.875 in the training cohort and 0.877 (95% CI 0.751–1.000), 0.833, and 0.826 in the validation cohort. The nomogram had good diagnostic performance, with AUCs of 0.925 (95% CI 0.862–0.988) and 0.899 (95% CI 0.782–1.000) in the training and validation cohort. Furthermore, the radiomics signature showed better diagnostic efficiency than the subjective assessment by both readers (AUC =0.912 vs. 0.732 and 0.763, P = 0.023 and 0.028, respectively). A radiomics nomogram was developed to preoperatively predict EMVI in RC patients. The application of the radiomics model based on relaxation maps could improve the diagnostic efficacy of EMVI.