Table 1.

ROC AUC and log loss results for the four different models as evaluated on the five different test sets.

Method	Source	ROC AUC	Log loss
two-stage $\times$ 10 224	Hospital 1	0.9633 [0.9335, 0.9864]	0.2892 [0.2251, 0.3396]
	Hospital 2	0.9590 [0.9365, 0.9771]	0.1931 [0.1413, 0.2461]
	Hospital 3	0.9903 [0.9771, 0.9983]	0.0651 [0.0435, 0.0948]
	Hospital 4	0.9669 [0.9209, 0.9972]	0.3075 [0.2121, 0.4048]
	Hospital 5	0.9932 [0.9863, 0.999]	0.1249 [0.0914, 0.1511]
one-stage $\times$ 20 512	Hospital 1	0.9835 [0.9776, 0.9973]	0.6243 [0.5222, 0.7274]
	Hospital 2	0.9696 [0.9375, 0.990]	0.3333 [0.2731, 0.400]
	Hospital 3	0.9862 [0.9824, 0.999]	0.3850 [0.3294, 0.4484]
	Hospital 4	0.9774 [0.9369, 0.9968]	1.1246 [0.9182, 1.3698]
	Hospital 5	0.9847 [0.9701, 0.9986]	0.5789 [0.512, 0.662]
one-stage $\times$ 20 224	Hospital 1	0.9594 [0.9314, 0.9818]	0.5874 [0.5324, 0.649]
	Hospital 2	0.9759 [0.9508, 0.993]	0.1988 [0.1810, 0.2202]
	Hospital 3	0.9751 [0.9464, 0.9944]	0.4177 [0.3942, 0.4501]
	Hospital 4	0.9714 [0.9281, 0.9935]	0.9354 [0.8142, 1.0714]
	Hospital 5	0.9774 [0.9498, 0.9978]	0.4428 [0.4128, 0.4682]
one-stage $\times$ 10 224	Hospital 1	0.8989 [0.8278, 0.9473]	1.4383 [1.2866, 1.6029]
	Hospital 2	0.9118 [0.8712, 0.9476]	0.2699 [0.2242, 0.3253]
	Hospital 3	0.8685 [0.7532, 0.9543]	1.0267 [0.9322, 1.1403]
	Hospital 4	0.8293 [0.6485, 0.9601]	1.8348 [1.5575, 2.1335]
	Hospital 5	0.9137 [0.8440, 0.9735]	1.1132 [1.0001, 1.2146]

First row group corresponds to the two-stage model, and the remaining three correspond to the one-stage models, with variations in magnification and tile size. Confidence intervals are between brackets.