Table 3.

Coefficients and odds ratios for individual features of the LungCNN-TMB model.

	Feature	Estimate (95% CI)	p-value	OR
Model coefficients	Solid	1.13 (0.49, 1.73)	0.001*	3.10
	Smoking status	0.90 (0.41, 1.36)	0.002*	2.46
	Age	−0.86 (−1.44, −0.29)	0.062	0.42
	Pathologic stage	−0.51 (−1.12, 0.09)	0.157	0.60
	Intercept	−1.56 (−1.90, −1.22)	0.000*	–

Two-sided P-values were estimated using bootstrapping. A logistic regression model combines features from the 9 image features of LungCNN-Histo and 4 clinical features (age in years, sex, pathologic stage, smoking status) to predict TMB status. The image features (quantified as proportion of total tumor area) and age (in years) are treated as continuous variables, pathologic stage (I–IV) as ordinal, and smoking status (binary) and sex (M,F) as categorical. Features in this table represent those providing best model via cross-validation (see “Methods”).

OR odds ratio