Geração de Regiões de Interesse Mamográficas utilizando um Modelo de Difusão Latente

Abstract

Breast cancer is the most common and leading cause of cancer de aths among women worldwide. In 2020, approximately 2.3 million new cases were registered in women, representing 11.7% of all cancer cases. Early diag nosis plays a crucial role in reducing breast cancer mortality. This study aims to generate Regions of Interest (ROIs) using a Latent Diffusion Model (LDM). The LDM is a probabilistic model that allows for the representation of complex latent information in images, which can be useful in identifying subtle patterns and efficiently generating ROIs. This work contributes to the advancement of synthetic medical image generation, offering a promising approach for genera ting diverse mammography ROIs. The effectiveness of the proposed method is evaluated based on the Mean Absolute Error, Mean Squared Error, and Frechet ´ Distance on a dataset of mammograms, from which the target regions for this application are extracted.