Comparative Analysis of Convolutional Neural Network Frameworks for Efficient Classification of Calcification Patches in Digital Breast Tomosynthesis Images

Breast cancer remains the most prevalent and rapidly spreading disease worldwide. Early detection significantly reduces mortality rates, driving the development of deep learning-based medical imaging analysis. This study focuses on classifying calcification patches in Digital Breast Tomosynthesis (DBT) images using Convolutional Neural Network (CNN) architectures. A comparative analysis is conducted on various CNN models, including GoogleNet, SqueezeNet, modified GoogleNet, and modified SqueezeNet, to evaluate their classification performance. The modified CNN models incorporate additional convolutional layers to enhance feature extraction for improved classification of DBT patches. Their performance is assessed using key evaluation metrics, including accuracy, precision, recall, and F1-score. Experimental results indicate that the modified GoogleNet achieves superior performance, attaining accuracy, precision, recall, and ${F 1}$-score values of ${9 7. 0 4 \%, ~} {9 1. 9 9 \%, ~} {9 6. 8 3 \%}$, and $94.23 \%$, respectively. Following classification of patches, positive patches undergo further processing for calcification localization using the You Only Look Once (YOLOv8) framework. This study underscores the efficacy of CNN-based approaches in classifying calcification patches in DBT images, demonstrating the potential of deep learning models in facilitating early breast cancer detection. The proposed methodology contributes to advancing medical imaging technologies, enhancing diagnostic accuracy, and improving patient outcomes.