Breast mass detection with kernelized supervised hashing

Abstract

In recent years, the number of breast cancer incidences has been growing continuously. Breast cancer has threatened female health seriously. As one of the main symptoms of breast cancer, mass detection is of paramount importance in computer-aided detection (CAD) systems. Breast mass detection has been widely applied to assist radiologists in locating abnormalities on mammograms. This paper presents a novel mass detection system for digital mammograms which can deal with various kinds of masses. A sliding window scheme is utilized to scan the breast area segmented from the whole mammogram. For every current window, Histogram of Oriented Gradient (HOG) is extracted and fed to a supervised algorithm, Kernel-Based Supervised Hashing (KSH), to obtain the corresponding compact binary code. In consideration of efficiency and accuracy, we propose a specific decision rule to classify the current window in hamming space. In order to label the detected mass region more accurately, a flexible sliding window fusion algorithm is proposed. Large scale experiments on Digital Database for Screening Mammography (DDSM) demonstrate the effectiveness and efficiency of the proposed detection scheme.