Three-Class Mammogram Classification Based on Descriptive CNN Features

Table 1

State-of-the-art diagnostic schemes for the screening mammography classification.

Authors	Year	Data sources	Technique/classifier	Classes	Number of images	Classification accuracy
Mazurowski et al. [3]	2011	DDMS	Random mutation hill climbing	2	1,852	49%–83%
Lesniak et al. [4]	2011	Private	SVM radial Kernel	2	10,397	66%-67%
Wei et al. [5]	2011	DDSM	SVM radial Kernel	2	2,563	72%–74%
Abirami et al. [7]	2016	MIAS	Wavelet features	2	322	93%
Tagliafico et al. [34]	2009	Private	Thresholding	4	160	80%–90%
Subashini et al. [35]	2010	Private	SVM radial Kernel	3	43	95%
Elter and Halmeyer [8]	2008	DDSM	Euclidean metric	2	360	86%
Deserno et al. [13]	2011	IRMA	SVM Gaussian Kernel	12	2796	80%
Tao et al. [6]	2011	Private	Local linear embedding metric	2	476	80%
			Curvature scale space		415	75%
Ge et al. [19]	2006	Private	CNN and LDA	2	196	—
		MIAS	CNN and LDA		216	—
Jamieson et al. [21]	2012	FFDM	ADN and SVM	2	739	—
		Ultrasound	ADN and SVM		2393	—
Arevalo et al. [22]	2015	BCDR-F03	CNN and SVM	2	736	79.9%–86%
Mert et al. [23]	2015	WBDC	ICA and RBFNN	2	569	90%
Dheeba et al. [25]	2015	Private	PSOWNN	2	216	93.6%
Abdel-Zaher and Eldeib [26]	2015	WBCD	DBN	2	690	99.6%
Vani et al. [10]	2010	MIAS	ELM
Jasmine et al. [11]	2009	MIAS	Wavelet & ANN	2	322	87%
Xu et al. [12]	2008		MLPNN		120	98%
Uppal and Naseem [27]	2016	MIAS	Fusion of cosine transform	3	322	96.97%