Investigation of various Supervised Machine Learning Algorithm to characterize Mammogram Breast Images

Abstract

Cancer is the prevalent global disease with a statistic of 68,500 deaths coupled with 2.3 million new cases in 2020 worldwide. Therefore, computer-aided early and accurate detection of disease is the prime concern for researchers and doctors. So, this work focuses on building a framework for accurate detection and diagnosis of mammogram image-based cancerous tissue. Different de-noising filters, such as anisotropic diffusion-based filter (ADF), median filter, wiener filter, and double wavelet transform (DWT) techniques are attempted for noise removal. The denoised imagesare segmented using the double thresholding segmentation technique and morphological area gradient-based segmentation (MAG). From the region of interest,grey level co-occurrence matrix (GLCM) features are extracted and analyzed. Results show, that ADF is the most efficient filter among the above-mentioned technique. Double thresholding segmentation is observed to segment breasts efficiently. Among the 11 GLCM features, four features discriminated between normal and abnormal subjects. The extracted features are subjected to supervised machine learning algorithms such as Decision Tree, Logical regression and Support vector machine. The quadratic support vector machine technique proved to be the most efficient among the compared. Hence, the proposed framework of ADF filter with double thresholding segmentation and GLCM feature shall be used for the characterization of mammogram breast images