Table 4.

Pixel accuracy (%) of the proposed augmentation according to the change in the number of nuclear items in an image used for training and testing.

Model	${\mathit{N}}_{\mathbf{test}}$	${\mathit{N}}_{\mathit{train}}=1$	${\mathit{N}}_{\mathit{train}}=2$	${\mathit{N}}_{\mathit{train}}=3$	${\mathit{N}}_{\mathit{train}}=4$
UNet	1	99.84	99.87	99.88	99.86
	2	99.61	99.76	99.79	99.77
	3	99.27	99.64	99.68	99.65
	4	98.83	99.45	99.53	99.49
	Average	99.39	99.68	99.72	99.69
FPN	1	99.76	99.78	99.95	99.82
	2	99.45	99.65	99.64	99.70
	3	99.08	99.49	99.49	99.58
	4	98.56	99.28	99.28	99.40
	Average	99.21	99.55	99.59	99.63
DeepLabV3+	1	99.82	99.79	99.82	99.82
	2	99.56	99.65	99.71	99.70
	3	99.16	99.49	99.57	99.57
	4	98.52	99.28	99.39	99.40
	Average	99.26	99.55	99.62	99.62
HRNet-OCR	1	99.83	99.80	99.80	99.77
	2	99.32	99.72	99.67	99.64
	3	98.80	99.64	99.52	99.49
	4	98.18	99.48	99.33	99.30
	Average	99.03	99.66	99.58	99.55