Pre-CADs in Breast Cancer
Joana Cristina, Lopes da Fonseca · 2013
In this study we present a pre-CAD system that aims to help the radiologists in the analysis of the high number of mammograms that they have to evaluate each day, helping to prevent the increased number of misclassification that could happen, due to the repetitive task to which they are submitted. The main objective of this work is to automate the classification of abnormal mammograms. As a pre-CAD system, not to miss any malignant mammogram is the most important point. The mammograms that the system was not sure, will be classified again by the medical staff. This system aims to help the medical staff in the mammograms that are difficult to analyse, providing an solid opinion based in the analysis of a trusted database. The method consists in extracting features from mammograms, previously classified by experts according to the breast density and then classify them into normal or abnormal mammograms for each of the tissue density types. A databased composed of 203 mammograms was used. Of this total number of mammograms, 153 were considered normal mammograms and 50 were considered abnormal mammograms. The 203 mammograms were classified by experts, according to the density of the breast, into dense and fatty, and three sets of mammograms were composed: a set composed by dense mammograms, another set composed by fatty mammograms and a set composed by the total number of mammograms. From each one of the sets, two types of features were extracted: Gray Level Co-occurrence Matrix features and Local Binary Pattern features. For the classification task, a crucial task in this work, three classifiers were used, in order to study the performance of each one of them. It was used the K-nearest neighbour, the Support Vector Machines and the Random Forests classifiers. The use, in this work, of three different scenarios allowed to study not only the performance of the classifiers, as well as the effect of the previous classification according to the density and the effect of the extraction of the different features.