Figure 2.

Texture feature selection using LASSO Cox regression model. (a) The partial likelihood deviation was plotted versus log (λ). The optimal parameter (λ) was selected using a sixfold cross-validation and 1 standard error rule. The optimal λ of 0.179 with log(λ) = −1.718 was selected. (b) LASSO coefficient profiles of 200 radiomics features were based on CT images. Finally, four features were selected, which are shown with black vertical dotted lines in the figure.