Fig. 2

Overall design for Experiment 2. a Feature extraction and the 1st selection step. In the 1st selection step, we selected features with ICC ≥ 0.7. In this process, we found that both histogram- and ISZM-based features have ICC ≥ 0.9. Thus, we fixed the histogram- and ISZM-based features to the default bin settings. b In the 2nd selection, we applied LASSO to select features that can explain nodule status. c The features were used to train a RF classifier to classify nodule status. It was later tested in a test cohort