Image-To-Image Translation Network Model Development for Automatic Contrast Enhancement on Low-Dose Mammographic Images

This study presents an Image-to-Image neural network model designed to enhance the contrast of low-dose mammographic images, aiming to improve diagnostic quality while adhering to the ALARA principle. Using a dataset of breast phantom images acquired at varying radiation doses, the model leverages generative adversarial networks (GANs) to reconstruct high-contrast images from low-dose scans. Performance evaluation based on structural similarity (SSIM) and Carneiro Contrast Index (CCI) demonstrates that the model effectively approximates high-dose image quality, supporting its potential role in reducing radiation exposure without compromising diagnostic accuracy.—