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

Table 16

Combination with SVM is compared with [23] (/%).

Accuracy	Location point	Magnifications
		40×	100×	200×	400×
Image-level	1	83.98 ± 2.90	84.99 ± 2.91	84.66 ± 2.41	82.78 ± 4.89
	2	87.85 ± 2.69	86.68 ± 2.28	87.75 ± 2.37	85.30 ± 4.41
	3	84.91 ± 1.58	84.60 ± 1.78	87.38 ± 3.52	84.78 ± 4.89
	fc6	83.00 ± 2.60	84.60 ± 5.00	84.00 ± 2.80	81.10 ± 3.90
	fc7	83.10 ± 2.10	83.30 ± 4.60	84.10 ± 1.50	81.60 ± 3.70
	fc8	83.20 ± 2.40	84.00 ± 4.90	83.40 ± 1.10	80.90 ± 3.70
Patient-level	1	81.93 ± 3.92	80.96 ± 3.61	82.40 ± 2.37	81.17 ± 3.04
	2	87.93 ± 3.91	87.41 ± 3.26	88.76 ± 2.50	85.55 ± 4.03
	3	83.91 ± 0.96	83.49 ± 2.65	87.89 ± 3.46	82.28 ± 4.11
	fc6	82.50 ± 8.60	83.60 ± 8.50	85.40 ± 5.20	81.10 ± 9.00
	fc7	83.40 ± 6.70	83.10 ± 8.40	86.00 ± 3.70	81.60 ± 8.60
	fc8	83.40 ± 6.90	83.80 ± 8.50	85.80 ± 3.50	80.70 ± 9.10

The fc6, fc7, and fc8 are the three top-most layers in [23].