Affiliations: [a] Centro de Investigación en Computación, Instituto Politécnico Nacional, 07700, CDMX | [b] Imagen Mamaria Centro Médico ABC | [c] Unidad Profesional Interdisciplinaria en Ingeniería y Tecnologías Avanzadas, Instituto Politécnico Nacional | [d] Centro de Innovación y Desarrollo Tecnológico en Cómputo, Instituto Politécnico Nacional
Correspondence:
[*]
Corresponding author. Hiram Calvo, Centro de Investigación en Computación, Instituto Politécnico Nacional, 07700, CDMX. E-mail: [email protected].
Abstract: In recent years, multiple studies have highlighted the growing correlation between breast density and the risk of developing breast cancer. In this research, the performance of two convolutional neural network architectures, VGG16 and VGG19, was evaluated for breast density classification across three distinct scenarios aimed to compare the masking effect on the models performance. These scenarios encompass both binary classification (fatty and dense) and multi-class classification based on the BI-RADS categorization, utilizing a subset of the ABC-Digital Mammography Dataset. In the first experiment, focusing on cases with no masses, VGG16 achieved an accuracy of 93.33% and 90.00% for two and four-class classification. The second experiment, which involved cases with benign masses, yielded a remarkable accuracy of 95.83% and 93.33% with VGG16, respectively. In the third and last experiment, an accuracy of 88.00% was obtained using VGG16 for the two-class classification, while VGG19 delivered an accuracy of 93.33% for the four-class classification. These findings underscore the potential of deep learning models in enhancing breast density classification, with implications for breast cancer risk assessment and early detection.
Keywords: Mammography, breast tissue density, convolutional neural networks
