Comparative study of dimensionality reduction techniques in mammographic images

Abstract

Breast cancer is a leading cause of cancer death among women. Early detection and diagnosis are essential for improving survival rates. In this study, we developed a computer-aided diagnosis (CAD) system based on machine learning approach for breast cancer detection. The proposed system is presented to investigate effects of dimension reduction techniques for classifying mammograms. It consists of preprocessing, feature extraction, feature selection, dimension reduction and classification steps. The Region of Interest (ROI) is extracted, and textural features are obtained using Local Binary Patterns (LBP). SelectKBest is employed as a feature selection technique to select the relevant features while Dimensionality reduction techniques, including Principal Component Analysis (PCA), Random Projection (RP), and Linear Discriminant Analysis (LDA), are applied to reduce the dimensionality of the images. K-Nearest Neighbors (KNN) and Support Vector Machines (SVM) with Sigmoid, Polynomial and Radial Basis Function kernels are used. The objective is to compare the performance of dimensionality reduction techniques and feature selection using these two classifiers. The initial results show that using classifiers alone did not achieve high accuracy. However, combining SelectKBest with PCA, RP, or LDA resulted in significant accuracy improvements. The best accuracy of 100% was achieved when combining SelectKBest with specific kernels of SVM and dimension reduction techniques. In conclusion, this research demonstrates the effectiveness of dimensionality reduction techniques, especially when combined with a feature selection technique which is SelectKBest in our case, in improving the classification accuracy for the detection of abnormalities in mammograms.