A Robust Feature Vector Based on Waveatom Transform for Mammographic Mass Detection
Wang Yang, Liu Tianhui · 2018
Breast cancer is one of the most common cancers diagnosed in women worldwide. Breast has different types of tissues, so they have textural variation in intensity. This makes diagnosis of abnormalities a challenging task. Therefore, there is a strong need to develop computer aided systems to act as a second opinion for radiologists. The main purpose of this research is to improve the accuracy of mammographic mass detection and reduce the computational complexity of the constructed feature vector. The proposed method is based on extracting normalized central moments from decomposed waveatom sub-bands. The feature vectors are investigated in terms of their capability to achieve the classification task using Random Forests with 10-fold cross validation. The reported results show an accuracy of 99.6% of mass vs. normal tissue classification. The results recommend the feature vector to be defined in terms of the second normalized central moment (Variance) computed for any sub-band of the first waveatom decomposition level.