Reconocimiento de la distorsión arquitectural en mamografías

Abstract

In this paper, we focus on the detection of
architectural distortion injury in mammography
images, based on a known methodology and
implementing improvements in its develoment. A
public mammography imaging database (MiniMIAS) is used, which contains the architectural
distortion lesion, among others. We show results
obtained using a Support Vector Machine Classifier
(SVM), with a vector of 5 features (of the 13
proposed by Haralick for texture analysis) and
selected through the Stepwise logistic algorithm
regression, to discern the benignity or malignancy
of the detected regions. Although a small number of
Mamograms with positive cases of architectural
distortion were used in this work, it is demonstrated
that the technique of support vector machines is
appropriate for these cases, offering an acceptable
performance.