Diagnóstico de Câncer de Mama: Melhoria na Classificação entre Lesões Benignas e Malignas com Radiômica

Abstract

Breast cancer is the most frequently diagnosed cancer among women and represents the leading cause of cancer-related death in this population. Early diagnosis is crucial for reducing mortality associated with this disease. However, detecting breast cancer at early stages poses challenges, particularly in patients with dense breast tissue. Ultrasonography is an effective tool that, when combined with mammography, aids in the identification and diagnosis of breast abnormalities. Moreover, radiomics seeks to extract, process, and clas sify imaging features, contributing to the differentiation between malignant and benign lesions. This study proposes the development of a Convolutional Neural Network (CNN) that leverages not only mammographic images but also their radiomic features to improve the accuracy of distinguishing malignant from be nign lesions. The results obtained with the final model, incorporating images along with their respective radiomic features, showed a 17% increase in accu racy and a 13% reduction in false-positive rates compared to a similar model using only images for classification.