Table 1.

Distributions of WSIs incorrectly predicted by patch-DNN models after classification by RRCART.

		Test set		High-accuracy set		Low-accuracy set
DNN-repeat-tumor normal patch ratio	Ratio (high/total)	Expected incorrectly predicted slides	P value	Expected incorrectly predicted slides	P value	Expected incorrectly predicted slides	P value	Fisher exact test for systematic error between high-accuracy and test set
InceptionV3-1-1:3	0.86	5.62	2.00E-06	4.17	7.86E-01	1.45	3.83E-13	0.003138
InceptionV3-1-1:5	0.81	5.83	9.27E-07	4.01	7.64E-01	1.81	1.28E-12	0.002974
InceptionV3-1-1:7	0.83	5.83	1.25E-04	4.25	9.25E-01	1.58	1.68E-10	0.003535
ReNet101-1-1:3	0.82	5.62	7.23E-06	4.26	7.98E-01	1.36	1.74E-12	0.006282
ReNet101-1-1:5	0.85	5.77	2.77E-09	4.68	6.87E-01	1.09	0.00E + 00	0.014472
ReNet101-1-1:7	0.83	5.73	5.24E-10	4.40	1.57E-01	1.33	0.00E + 00	0.095278
Xception-1-1:3	0.83	5.94	5.00E-05	4.38	4.45E-01	1.56	1.20E-08	0.007071
Xception-1-1:5	0.85	5.77	7.97E-07	4.25	4.19E-01	1.52	9.76E-12	0.009791
Xception-1-1:7	0.84	5.77	3.49E-05	4.43	2.85E-01	1.34	2.12E-08	0.037669
InceptionV3-2-1:3	0.84	5.83	3.45E-06	4.14	5.92E-01	1.70	4.66E-11	0.01299
InceptionV3-2-1:5	0.83	5.89	4.52E-05	4.32	4.34E-01	1.57	1.29E-08	0.036724
InceptionV3-2-1:7	0.83	5.83	1.06E-03	4.50	8.27E-01	1.33	1.81E-08	0.013734
ReNet101-2-1:3	0.81	5.81	3.83E-05	4.40	8.15E-01	1.41	3.69E-11	0.008692
ReNet101-2-1:5	0.83	5.65	2.14E-06	4.29	6.21E-01	1.35	1.72E-12	0.00309
ReNet101-2-1:7	0.81	5.93	3.29E-03	4.18	2.44E-01	1.75	4.63E-04	0.147115
Xception-2-1:3	0.82	5.89	1.37E-05	4.40	6.41E-01	1.49	7.46E-11	0.014449
Xception-2-1:5	0.84	5.81	8.92E-07	4.08	3.88E-01	1.73	5.96E-11	0.022463
Xception-2-1:7	0.81	5.63	2.53E-05	4.44	9.36E-01	1.19	2.54E-13	0.002071