A Convolutional Neural Network Model for Image Enhancement of Extremely Dense Breast Tissue in Digital Breast Tomosynthesis Images

Abstract

Almost half of all mammograms performed on women over 40 reveal dense breasts. Breast density has been associated with both low BMI and the use of postmenopausal hormone replacement therapy. Breast cancer is more common in women with dense breasts than in women with fatty breasts, and the risk increases with increasing breast density. Mammograms may be more challenging to read in women with dense breasts than in women with fatty breasts. This is because dense breast tissue and pathological breast alterations, such as calcifications and tumours, appear as white regions on mammography. As a result, this study aims to create a fast and accurate approach for enhancing Digital Breast Tomosynthesis (DBT) images using a deep learning convolution neural network (CNN). The results then show that the very deep super-resolution (VDSR) approach produces the best results compared to the BICUBIC. Finally, this research will compare the obtained results for VDSR with the traditional method, BICUBIC. This study can conclude that the project successfully obtained a satisfactory result.