Abstract

A support vector machine (SVM) classifier is used to determine whether regions of interest (ROIs) found on breast radiographic images contain mass or normal tissue. Before being presented to SVM, ROIs are encoded by means of a specific image representation. The coefficients resulting from the encoding are then used as classification features. Pixel and wavelet image representations have already been discussed in one of our previous works. A novel orientation–selective, non–parametric and multi–resolution image representation is developed and evaluated herein, namely a ranklet image representation. From the digital database for screening mammography (DDSM) collected by the University of South Florida, a database of ROIs is generated. A total of 1000 ROIs containing diagnosed masses are extracted from the DDSM benign and malignant cases, whereas 5000 ROIs containing normal tissue are extracted from the DDSM normal cases. The area Az under the receiver operating characteristic curve is adopted for performance evaluation. By achieving Az values of 0.978 ± 0.003, experiments demonstrate better classification results with respect to those reached by the previous image representations. In particular, the improvement on the Az value over that achieved by the wavelet image representations is statistically relevant with two–tailed p–value < 0.0001.