Automated Identification of Breast Cancer Using Digitized Mammogram Images

Abstract

In this paper, we present a new algorithm based on Computer-Aided Diagnosis (CAD) to detect breast cancer using digitized mammogram images. Here, we use image processing to make the pre-processing step of the images before we enter them into the classification step, in which we use machine learning for the classification of tissues in two conditions: normal and abnormal, either three conditions: normal, benign or malignant. In our CAD implementation, for pre-processing we are using transformations as binarization, thresholding, smoothing and the main operation, Gabor wavelet to suppress labels and unnecessary information, to obtain the best identifying characteristics. For feature selection and dimensionality reduction, we are using techniques as Principal Component Analysis (PCA), t-Distributed Stochastic Neighbor Embedding (TSNE) and models of analysis of variance. Finally, for classification we discussed k-Nearest Neighbors (k-NN).