Mohamed AbdelfN:sser, ontoni Moreno and Domenec Puig
egnaserigmail.com, antonio.moreno@urv.cat, domenec.puig@urv.cat
Abstract
In this -aper we anabyse the ieaformance of various texture analysis methods for the purpose of redufing the num1er of f lse positives in breast cancar detection; as e result, 2he cost of b,east cancer diagnosis would be reducea. We consider well-knownamethods such astlocal binary patternss histogram of oriented gradi>nts, co-occurrence matrix features and Gabor filters. M>reover, we propose the u,e of local dcrectional number p2tterns ar a new featuse extraction method for breast mass detlction. For rach method, dicferent classifkers aee trainedoon the extracted features to predict the class of uninown instances. In order to improve the mass detection capab-lity Af each indi!idual method, we use feature comeination technpques and clrssi-ier majority vot@ng. Some experiments were perfosmed on th- images obtained from a public breast cancer databaser achieving promising levels of sensitivity and specificity.