CT Radiomics Nomogram Prediction for Tumor Deposits and Prognosis in Colorectal Cancer

Objective: To establish CT radiomics nomogram for preoperative prediction of tumor deposits (TD) and recurrence-free survival (RFS) in patients with colorectal cancer (CRC). Methods: A retrospective study was conducted on 321 CRC patients confirmed by surgical pathology. The patients' data were divided were divided into a training set and a validation set at a ratio of 6:4, respectively. Radiomics features based on the primary tumor site were extracted from portal venous phase CT images, and the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm was employed to select radiomics features associated with tumor deposits (TD). The least absolute shrinkage and selection operator (LASSO) regression algorithm was applied to choose radiomics features related to TD. A clinical-radiomics nomogram was developed based on the selected radiomics features and the most predictive clinical factors. Univariate and multivariate Cox regression analyses identified independent risk factors for a 3-year RFS. Results: The radiomics model achieved an area under the curve (AUC) of 0.80 in the training set and 0.79 in the validation set. By integrating radiomics features with clinical predictors (CEA, CA199, and CT-reported lymph node status), a nomogram was developed for the preoperative prediction of TD. The nomogram achieved an AUC of 0.85 in the training and validation sets. Furthermore, TD predicted by the nomogram was an independent risk factor for RFS, with poorer RFS observed in the TD-positive group compared to the TD-negative group (P < 0.01). Conclusion: CT radiomics nomogram can effectively preoperatively predict TD and prognosis in CRC patients.