A New GLLD Operator for Mass Detection in Digital Mammograms

Table 1

Previously developed approaches on mass detection based on feature extraction and on learning. In this table, we specify for each approach the feature extraction technique, the classifier, the ratio which indicates the number of real masses/number of normal ROIs, and the obtained results. In the feature extraction methods, ICA, PCA, and 2DPCA correspond, respectively, to independent component analysis, principal component analysis and two-dimensional PCA. In the classification stage, ANN, NN, and SVM correspond, respectively, to the artificial neural network, nearest neighbors, and support vector machines. Generally, the evaluation of the works is given in terms of 𝐴_𝑧 where 𝐴_𝑧 represents the area under the ROC curve, except for both works of Christoyianni et al. and Leonardo et al. giving the correct classification true positive and true negative in percentage.

Classifier based
Author	Year	Texture	Morphology	Shape	Gray level	ICA	PCA	2DPCA	Classifier	Ratios	Results
Qian et al. [6]	2001	√		√					ANN	200/600	𝐴𝑧 = 0.86
Christoyianni et al. [7]	2002	√			√		√		ANN	119/119	88.23%
Oliver et al. [8]	2006						√		C4.5 + NN	196/392	𝐴𝑧 = 0.83
Oliver et al. [9]	2007							√	NN	256/1536	𝐴𝑧 = 0.86
Varela et al. [10]	2007		√		√				ANN	60/60	𝐴𝑧 = 0.90
Leonardo et al. [11]	2009	√		√					SVM	250/1177	92.63%