Table 2.

Performance on the whole-slide level on the independent test sets.

Test set	Point	A-Basic				B-Rescanned				C-Color norm.				D-GAN style norm.
		SE	SP	ACC	AUC	SE	SP	ACC	AUC	SE	SP	ACC	AUC	SE	SP	ACC	AUC
IT1	P1	1	0.41	0.81	0.92 ± 0.03	1	0.3	0.78	0.91 ± 0.04	1	0.72	0.9	0.96 ± 0.03	1	0.5	0.84	0.98 ± 0.01
	P2	0.64	1	0.75		0.42	1	0.60		0.74	1	0.82		0.74	1	0.86
	P3	0.92	0.65	0.83		0.93	0.73	0.87		0.93	0.87	0.91		0.96	0.87	0.93
IT2	P1	1	0.05	0.66	0.83 ± 0.06	1	0.11	0.68	0.88 ± 0.05	1	0.05	0.65	0.81 ± 0.07	1	0.75	0.91	0.97 ± 0.03
	P2	0.33	1	0.57		0.53	1	0.7		0.21	1	0.49		0.69	1	0.80
	P3	0.88	0.53	0.75		0.85	0.75	0.81		0.91	0.50	0.76		0.99	0.80	0.92

Results are presented for three points (thresholds) of the ROC curve: P1—specificity equal 1, P2—sensitivity equal 1, P3—the best accuracy, where: A—original slides, B—rescanned slides, C—color normalized slides, D—style normalized slides by cycle-GAN method, SE—sensitivity, SP—specificity, ACC—accuracy