Classification on Digital Pathological Images of Breast Cancer Based on Deep Features of Different Levels

Table 9

Image-level accuracy and patient-level accuracy evaluation results at different magnifications (/%).

Accuracy	Strategy	Magnifications
		40×	100×	200×	400×
Image-level	Sliding	84.19 ± 2.27	83.80 ± 2.60	86.48 ± 2.80	83.93 ± 3.93
	Random	82.54 ± 2.81	84.41 ± 1.53	86.72 ± 2.77	82.74 ± 3.16
	Sliding + Random	84.50 ± 1.26	84.70 ± 1.25	86.95 ± 1.52	83.25 ± 2.47
	Sliding + class Balance Random	85.04 ± 0.87	85.66 ± 1.07	86.97 ± 0.77	84.29 ± 2.36
Patient-level	Sliding	85.70 ± 1.90	83.93 ± 2.12	87.73 ± 3.32	83.70 ± 4.04
	Random	83.73 ± 1.76	84.46 ± 2.52	88.31 ± 2.47	82.75 ± 3.35
	Sliding + Random	86.05 ± 2.17	86.03 ± 1.92	88.18 ± 3.31	83.31 ± 2.43
	Sliding + Class Balance Random	86.06 ± 1.62	86.29 ± 2.85	88.69 ± 4.17	83.57 ± 5.04

The optimal value of the evaluation result has been bolded.