MRI-based multiregional radiomics for preoperative prediction of tumor deposit and prognosis in resectable rectal cancer: a bicenter study

ObjectiveTo build T2WI-based multiregional radiomics for predicting tumor deposit (TD) and prognosis in patients with resectable rectal cancer.Materials and methodsA total of 208 patients with pathologically confirmed rectal cancer from two hospitals were prospectively enrolled. Intra- and peritumoral features were extracted separately from T2WI images and the least absolute shrinkage and selection operator was used to screen the most valuable radiomics features. Clinical-radiomics nomogram was developed by radiomics signatures and the most predictive clinical parameters. Prognostic model for 3-year recurrence-free survival (RFS) was constructed using univariate and multivariate Cox analysis.ResultsFor TD, the area under the receiver operating characteristic curve (AUC) for intratumoral radiomics model was 0.956, 0.823, and 0.860 in the training cohort, test cohort, and external validation cohort, respectively. AUC for the peritumoral radiomics model was 0.929, 0.906, and 0.773 in the training cohort, test cohort, and external validation cohort, respectively. The AUC for combined intra- and peritumoral radiomics model was 0.976, 0.918, and 0.874 in the training cohort, test cohort, and external validation cohort, respectively. The AUC for clinical-radiomics nomogram was 0.989, 0.777, and 0.870 in the training cohort, test cohort, and external validation cohort, respectively. The prognostic model constructed by combining intra- and peritumoral radiomics signature score (radscore)-based TD and MRI-reported lymph nodes metastasis (LNM) indicated good performance for predicting 3-year RFS, with AUC of 0.824, 0.865, and 0.738 in the training cohort, test cohort and external validation cohort, respectively.ConclusionCombined intra- and peritumoral radiomics model showed good performance for predicting TD. Combining intra- and peritumoral radscore-based TD and MRI-reported LNM indicated the recurrence risk.

Automated summary

The study aimed to develop a T2WI-based multiregional radiomics model for predicting tumor deposit (TD) and prognosis in patients with resectable rectal cancer. Intra- and peritumoral features were extracted from T2WI images and the most valuable radiomics features were determined. A clinical-radiomics nomogram and a prognostic model for 3-year recurrence-free survival (RFS) were constructed using these features. The models showed good performance in predicting TD and recurrence risk.