TABLE 3.

Performance of ResNet50, InceptionV3 and InceptionResNetV2 models with or without fine grained methods on images with different magnification factors.

		Accuracy (%)
		Magnification factor
Backbone network	Fine-grained method	40×	100×	200×	400×	Mixed
Inceptionv3	Not used	91.72 ± 0.16	91.84 ± 0.22	89.83 ± 0.17	87.64 ± 0.22	92.24 ± 0.13
	SE Block	94.99 ± 0.31	94.48 ± 0.24	94.79 ± 0.26	94.23 ± 0.39	94.90 ± 0.23
	BCNNs	95.74 ± 0.21	94.72 ± 0.18	94.78 ± 0.26	94.51 ± 0.23	96.14 ± 0.16
ResNet50	Not used	90.23 ± 0.15	90.17 ± 0.32	89.83 ± 0.54	89.01 ± 0.48	92.48 ± 0.29
	SE Block	92.98 ± 0.32	90.93 ± 0.26	88.09 ± 0.38	92.31 ± 0.28	92.79 ± 0.22
	BCNNs	95.74 ± 0.24	95.44 ± 0.32	94.53 ± 0.25	94.88 ± 0.34	95.27 ± 0.26
InceptionResNetV2	Not used	91.48 ± 0.12	91.52 ± 0.19	92.30 ± 0.26	90.65 ± 0.26	92.62 ± 0.17
	SE Block	94.99 ± 0.28	93.93 ± 0.34	92.31 ± 0.29	91.97 ± 0.36	94.02 ± 0.30
	BCNNs	95.24 ± 0.13	94.16 ± 0.22	94.03 ± 0.29	93.70 ± 0.23	95.08 ± 0.14

Average accuracies ± standard deviations were calculated from 5-fold cross validation of the BreaKHis dataset.