Table 1. Performance metrics of each set of features for the Opacity Score and Geographic Extent prediction.

Evaluation is performed on 50 randomly chosen train test splits and the metrics here are computed on a hold out test set.

R2: coefficient of determination; MAE: mean absolute error; MSE: mean squared error. “four outputs” refers to lung opacity, pneumonia, infiltration, and consolidation

Task	Using features:	# parameters (fewer is better)	Pearson Correlation	R2	MAE	MSE
Opacity Score	"lung opacity" output	1+1	0.78±0.04	0.58±0.09	0.78±0.05	0.86±0.11
	4 outputs	4+1	0.78±0.04	0.58±0.09	0.76±0.05	0.87±0.12
	18 outputs	18+1	0.73±0.09	0.44±0.16	0.86±0.11	1.15±0.33
	Intermediate network features	1024+1	0.66±0.08	0.25±0.21	1.01±0.09	1.54±0.28
	No data	0+1	-0.00±0.00	-0.08±0.10	1.24±0.10	2.26±0.36
Geographic Extent	"lung opacity" output	1+1	0.80±0.05	0.60±0.09	1.14±0.11	2.06±0.34
	4 outputs	4+1	0.79±0.05	0.57±0.10	1.19±0.11	2.17±0.37
	18 outputs	18+1	0.76±0.08	0.47±0.16	1.32±0.17	2.73±0.89
	Intermediate network features	1024+1	0.74±0.08	0.43±0.16	1.36±0.13	2.88±0.58
	No data	0+1	0.00±0.00	-0.08±0.10	2.00±0.17	5.60±0.95