Exploring Deep Learning and Transfer Learning for Colonic Polyp Classification

Table 8

Accuracies of the methods for the CC-i-Scan databases in %.

Methods	No staining				Staining
	CVC	i-Scan1	i-Scan2	i-Scan3	CVC	i-Scan1	i-Scan2	i-Scan3
1: CNN-F	86.16	89.33	80.65	88.41	86.52	81.40	84.22	80.62	84.66
2: CNN-M	87.45	90.67	81.38	83.58	87.99	89.55	87.40	90.53	87.31
3: CNN-S	88.03	90.00	87.01	77.33	87.25	82.68	87.40	75.54	84.41
4: CNN-F MCN	88.84	82.00	73.15	90.73	85.78	89.55	89.72	83.15	85.36
5: CNN-M MCN	89.53	90.67	88.88	94.66	86.97	89.29	87.40	90.53	89.74
6: CNN-S MCN	90.12	91.42	81.38	79.85	89.18	93.49	81.10	84.77	86.41
7: GoogleLeNet	79.65	90.67	72.43	74.51	88.27	80.46	75.60	84.08	80.70
8: VGG-VD16	87.45	85.33	86.38	79.65	92.47	89.80	95.26	92.38	88.59
9: VGG-VD19	83.49	82.67	83.88	87.71	92.47	83.98	94.46	85.59	86.78
10: AlexNet	91.40	87.33	75.65	89.32	87.71	83.03	84.22	79.24	84.73
11: AlexNet MCN	89.42	84.67	78.88	83.78	89.36	83.55	81.10	78.32	83.63
	87.41	87.70	80.88	84.50	88.54	86.07	86.17	84.06	85.67
12: BSAG-LFD	86.27	86.87	84.60	82.87	70.20	80.63	78.78	71.39	80.20
13: Blob SC	77.67	83.33	82.10	75.22	59.28	78.83	66.13	59.83	72.79
14: Shearlet-Weibull	73.72	76.67	79.60	86.80	81.30	69.91	72.38	83.63	78.00
15: GWT-Weibull	79.75	78.67	70.25	84.28	81.30	74.54	77.17	83.39	78.66
16: LCVP	76.60	66.00	47.75	77.12	77.45	79.00	70.01	69.56	70.43
17: MB-LBP	78.26	80.67	81.38	83.37	69.29	70.60	77.22	78.32	77.38
	78.71	78.70	74.28	81.61	73.13	75.58	73.61	74.35	76.24
Fusion 5/8	88.84	85.33	83.88	92.14	93.12	90.49	96.88	94.00	90.58
Fusion 5/12	92.79	92.67	88.88	96.98	87.71	90.49	88.26	90.53	91.03
Fusion 5/8/12	95.94	90.00	88.88	92.14	92.30	91.43	97.63	97.46	93.22
Fusion 5/8/14	91.51	88.67	87.10	93.75	94.68	91.43	98.44	95.85	92.67
Fusion 5/8/15	90.91	90.00	88.88	92.14	93.94	89.80	96.88	95.61	92.27
Fusion 5/8/12/14	93.38	88.00	91.38	93.75	93.49	92.12	97.63	94.92	93.08
Fusion 5/8/12/17	93.38	90.00	91.38	93.75	92.75	92.12	97.63	97.46	93.55
CNN-05	—	91.00	—	89.00	—	—	—	—	—
CNN-05 + SVM	—	83.00	—	72.55	—	—	—	—	—