Table 2.

LASSO-regularized logistic regression performances in HER2+ and TNBC cohorts.

Cohort	Features	AUC	F1 score	Precision (PPV)	Recall	NPV
HER2+	IMPRESS (all features)	0.8975 ± 0.0038	0.8687 ± 0.0077	0.8716 ± 0.0115	0.8658 ± 0.0081	0.7897 ± 0.0110
	IMPRESS (H&E only)	0.8118 ± 0.0048	0.8269 ± 0.0052	0.9059 ± 0.0009	0.7605 ± 0.0081	0.6977 ± 0.0070
	IMPRESS (IHC only)	0.7746 ± 0.0057	0.7775 ± 0.0085	0.8454 ± 0.0023	0.7197 ± 0.0129	0.6399 ± 0.0105
	Pathologists’ features	0.7880 ± 0.0065	0.7820 ± 0.0025	0.8696 ± 0.0061	0.7105 ± 0.0000	0.6446 ± 0.0026
TNBC	IMPRESS (all features)	0.7674 ± 0.0209	0.7017 ± 0.0377	0.6903 ± 0.0286	0.7148 ± 0.0552	0.7714 ± 0.0344
	IMPRESS (H&E only)	0.6795 ± 0.0103	0.5882 ± 0.0000	0.6250 ± 0.0000	0.5556 ± 0.0000	0.7000 ± 0.0000
	IMPRESS (IHC only)	0.5975 ± 0.0087	0.5915 ± 0.0103	0.5637 ± 0.0061	0.6222 ± 0.0152	0.7018 ± 0.0085
	Pathologists’ features	0.7626 ± 0.0095	0.6897 ± 0.0077	0.6454 ± 0.0135	0.7407 ± 0.0000	0.7878 ± 0.0042

PPV positive predictive value, NPV negative predictive value.

Experiments are repeated 20 times with different random seeds in leave-one-out cross-validation setting. mean value ± standard deviation are reported. Best performed mean values are highlighted in bold face.