Effect of image binarization thresholds on breast cancer identification in mammography images using OTSU, Niblack, Burnsen, Thepade's SBTC

Abstract

Mortality rates due to breast cancer in women are very high. Accurate prediction and detection at an early stage of cancer are required for a reduction in death rates. A Computer-Aided Diagnosis (CAD) is very useful for doctors in deciding screening levels. For identifying cancerous tumors, a CAD-based classification of a benign, malignant tumor using Machine learning is one of the approaches. This paper has observed the effect of various thresholding methods like Otsu, Niblack, Bernsen, Thepade's Sorted block Truncation Coding (TSBTC), and feature level fusions of them in the identification of breast cancer at an early stage. Extracted features using thresholding techniques are fed as an input to machine learning algorithms for the classification of Benign, malignant, normal tumors, which leads to the identification of breast cancer. For experimentation purposes, a Mammographic Image Analysis Society (MIAS) dataset for Breast cancer detection is used. The accuracy of these machine learning algorithms is calculated using Weka's open-source machine learning software. Finally, the accuracy of machine learning classifiers along with presented thresholding techniques is evaluated.