Figure 3.

Feature selection using the LASSO model. (A) In the primary dataset, the penalization parameter λ was selected using 10-fold cross-validation LASSO method with the mean squared error (MSE) as the criterion. In this study, the minimum MSE was at λ = 0.069, log(λ) = –2.68. (B) LASSO coefficient profile of 129 radiomic features. There are 23 nonzero coefficient features at the optimal λ. LASSO, least absolute shrinkage and selection operator.