Segmentation of Noisy Mammograms Using Hybrid Techniques

Abstract

Breast cancer is the most routinely identified carcinoma among women in India, and it is one of the foremost causes of cancer death in women. Radiologists prefer mammograms for visualizing breast cancer. Different types of noises including Gaussian noise and salt-and-pepper noise affect the mammograms leading to inaccurate classification. Mammograms consist of numerous artifacts too, which depressingly affect the finding of breast cancer. The existence of pectoral muscles makes anomaly finding a cumbersome task. The recognition of glandular tissue in mammograms is vital in assessing asymmetry between left and right breasts and in conjecturing the radiation risk associated with screening. Thus, the proposed method focuses on improving the segmentation accuracy of noisy mammograms. It involves preprocessing which includes denoising using a pretrained convolutional neural network, artifacts removal using thresholding, and modified region growing and enhancement using two-stage adaptive histogram equalization along with segmentation of mammogram images into sections conforming to different densities using K-means clustering. The projected method has been confirmed on the Mini-MIAS database with ground truth provided by expert radiologists. The results illustrate that the proposed method is competent in eradicating noise and pectoral muscles without degrading quality and contrast and in fragmenting different denoised mammograms into different mammographic densities with high accuracy.