Fig. 3
Construction of immune scoring system by integrating multiple machine learning methods. A–C The importance feature maps show the top 10 important features in LASSO model, XGBoost model and Random Forest model. D Venn diagram demonstrates the six immune features shared by the three models: density of CD103+ cells in TC, density of TRM3 in IM, infiltration score of Tnon-RM in TC, infiltration score of TRM1 in IM, infiltration score of TRM4 in IM and mean nearest neighboring distance from TRM to cancer cells in TC. E Univariate Cox analyses show the prognostic relevance of the six selected features. F–H Time-dependent ROC curves and AUC values of TRM-SIS system and TNM staging system for prediction of recurrence risk at 1, 3 and 5 years in the training set, testing set and entire cohort. The TRM-SIS system presents better performance than the TNM staging system for recurrence prediction. I–K Survival analyses based on the TRM-SIS in the training set, testing set and entire cohort. Statistical significance was calculated using the log-rank test. AUC, area under the curve; DFS, disease-free survival; ROC, receiver operating characteristic