Detecting breast cancer in mammograms can be a hard task, particularly
if the patient has dense breasts. Several works in the literature have
tried to build models to describe malignant or benign findings using
BI-RADS annotated features or features automatically extracted from
images. Some of the best models are based on Support Vector Machines
(SVMs). Features from mammograms have heterogeneous types and SVMs
handle them equally. Multiple Kernel Learning (MKL) can learn models
where each feature can be treated in a different way, which may
improve the quality of the models. In this work, we apply MKL with SVM
in tree diferent datasets to prove that MKL can get bether results than
SVM when dealing with the same data.
For BCDR data, MKL and CSMKL got worst results than SVM at 0.5 threshold,
CSMKl or MKL get better results when we want to achieve 100% Recall for
the Malignant class. For the 2 DDSM datasets we did not use CSMKL.
For the first dataset o SVM only got 0.57 accuracy which means that
it was practically random. With MKL we got 62% accuracy . For second
dataset the SVM only got 58% accuracy and MKL got 78% accuracy.
keywords: SVM, MKL, Data Mining, Breast Cancer