Fig. 2
From: Predicting response to CCRT for esophageal squamous carcinoma by a radiomics-clinical SHAP model
Lasso feature filtering. Selection of the best penalty coefficient value (lambda) that has the minimum mean squared error (red dot) (MSE) in tenfold cross-validation of LASSO regression for CTV (a) and GTV (b) features. The optimal lamda value (CTV: 0.02, GTV:0.04, dashed vertical line) was chosen as the smallest mean MSE of the training sample. Lasso coefficient profiles (c, d) against lambda value and the best lambda value (dashed vertical line) where retained features had non-zero coefficients