Fig. 1

Radiomic feature selection. Performed in a training set (67% of the participants). Max and min {AUCt} denote the maximum and minimum values of the time-dependent area under curve, respectively, across 12 time cutoffs ranging 1–12 months (defined as 30.5–366 days). max{AUCt} ≥ 0.7 indicates high predictive accuracy of lung cancer, and min {AUCt} ≥ 0.6 indicates stable predictive accuracy over time. ICC denotes the intraclass coefficient between feature values extracted from the original images and those extracted from noised images. ICC < 0.8 indicates non-robustness of the radiomic feature. *The 17 features were categorized into 6 groups, within which the features are highly correlated (pairwise Spearman r > 0.80). VIF denotes the variance inflation factor. VIF < 10 indicates a lack of collinearity between the finally selected features (i.e., that they are independent characterizations of the nodule)