Content-based image retrieval applied to BI-RADS tissue classification in screening mammography

Table 1 Parameters of the kernels of the support vector machine model

Parameters	Polynomial kernel	Radial kernel
Cost (C)	10	1
Gamma (g)	0.00022	0.0055
Epsilon (e)	0.1	0.1
Degree	2	-

Table 2 Average precision results of the MammoSVx system

k	Feature	Average precision (mean ± SD, %)			Trace of Ak
		Polynomial	Radial	Linear
25	SVD	79.42 ± 0.44	68.92 ± 0.05	75.36 ± 0.10	23
	SVD + histogram	82.14 ± 0.30	70.48 ± 0.12	75.66 ± 0.01
50	SVD	76.61 ± 0.40	71.00 ± 0.09	75.12 ± 0.07	24
	SVD + histogram	76.72 ± 0.50	71.47 ± 0.16	76.26 ± 0.04
100	SVD	77.91 ± 0.15	69.23 ± 0.16	75.32 ± 0.06	24
	SVD + histogram	78.31 ± 0.14	71.75 ± 0.04	75.70 ± 0.18
150	SVD	76.85 ± 0.16	71.52 ± 0.06	74.41 ± 0.14	25
	SVD + histogram	76.47 ± 0.20	68.76 ± 0.17	76.44 ± 0.16
200	SVD	75.69 ± 0.16	71.57 ± 0.06	74.80 ± 0.08	26
	SVD + histogram	76.67 ± 0.23	70.87 ± 0.16	76.87 ± 0.04
	Histogram only	50.00	50.00	67.80

SVD: Singular value decomposition.