Fig. 3

Features selection for the prediction models by LASSO regression. Tuning parameter (λ) selection used 10-folds cross-validation. The X-axis shows log (λ), and the Y-axis shows the model misclassification rate. The 2, 3, 3,4 features with non-zero coefficients are indicated with the optimal λ values of 0.07, 0.10, 0.08, 0.10 for Clinical model (a), CT model (b), TA model (c), Combined model (d), respectively