Prediction of Lymph Node Metastasis by Intratumoral and Peritumoral CT Radiomics Combined with Clinical Features in Rectal Cancer

DONG Li; GONG Xutao; LI Jing; LIU Ning

doi:10.3969/j.issn.2096-305X.2025.04.010

Abstract：Objective This study aims to establish a machine learning model based on conventional CT radiomics features to predict the preoperative lymph node metastasis(LNM)status of rectal cancer and evaluate the gain value of peritumoral quantitative features.Methods A retrospective study included 371 patients with pathologically confirmed rectal cancer who underwent CT examinations at Pingyi County Traditional Chinese Medicine Hospital from January 2019 to December 2023.The patients were randomly divided into the training set and the test set in a ratio of 7∶3.6,264 radiomics features were extracted from the intratumoral and peritumoral(3,5,7 mm)areas of CT.After feature screening,intratumoral and peritumoral radiomics machine learning prediction models were constructed respectively.In addition,clinical models and combined models were established in combination with clinical factors.The performance of each model in predicting LNM was analyzed and compared using the receiver operating characteristic(ROC)curve,and indicators such as the area under the curve(AUC),sensitivity,and specificity were calculated.And further construct the nomogram of the best model.Results Among the enrolled patients,144 cases(38.81％)had lymph node metastasis.In the test set,the model based on the radiomics characteristics of 5 mm around the tumor had the best predictive efficacy(AUC=0.810),and was significantly better than the intratumoral model(AUC=0.709,P=0.046).The comprehensive model constructed by combining the best intratumoral and peritumoral models with clinical factors achieved AUC values of 0.871 and 0.890 respectively in the training set and test set,which was significantly better than the single clinical model(AUC=0.736 and 0.710,P＜0.05),and had the highest clinical benefit.Conclusion The combined model that integrates the intratumoral and peritumoral radiomics features of CT and combines clinical features has shown excellent efficacy in predicting the preoperative lymph node metastasis status of rectal cancer.The quantitative characteristics around the tumor may reflect the unique biological process of metastasis and are of great value for improving the non-invasive prediction of lymph node metastasis status of rectal cancer.The nomogram constructed based on the joint model is expected to provide strong support for clinical individualized diagnosis and treatment decisions.