Fig. 1
From: Predicting response to CCRT for esophageal squamous carcinoma by a radiomics-clinical SHAP model
Overall pipeline of this work. a Pre-CCRT data acquisition and delineations. Cases were screened according to patient inclusion criteria. ROIs were outlined by experienced radio-oncologists. Features were extracted using PyRadiomics from ROIs. b Feature selection. Clinical, CTV and GTV features were filtered by statistical analysis, lasso and mutual information in sequence as described in the feature selection section. c Construction and comparison of five predictive models with random forest classifiers and filtered features including CTV, CTV-Clinical, Clinical, GTV and GTV-Clinical models. d Model interpretation at the global and local levels with SHAP algorithm