Identifying an Efficient Denoising Method for Threshold-based Segmentation of Mammograms
Ranjan K. Pradhan, Pralipta Mallick, Satabdi Mishra · 2024
Mammography has been the most widely used imaging modality to diagnose breast cancer and other abnormalities in breast tissues. Denoising of mammography image is a fundamental step for efficient detection of breast tumor, in which the foreground breast tissue is segmented out of the 2D-image. Although several methods have been proposed to remove unwanted noises and pectoral muscles from mammogram images, the efficacies of classical denoising filters on improving the threshold-based segmentation accuracy in mediolateral oblique (MLO) view of mammogram is unclear. In this study, we have systematically evaluated the effects of three most widely used image denoising filters on threshold-based segmentation of MLO-mammogram, obtained from MIAS database. A novel preprocessing framework was proposed and validated for effective denoising, image enhancement and pectoral muscle detection in mammograms of dense, glandular and dense-glandular breast masses using a robust global thresholding method. The performances of denoising filters were evaluated based on the computed values of their peak noise to signal ratio (PSNR) and mean squared error (MSE) for a total 320 images from MIAS database. These results suggest, the Wiener filter-based denoising has remarkable effect on improving image quality, and seems more suitable for global thresholding-based segmentation. The present method provides a simple, versatile and effective way of preprocessing digital mammograms.