BreNet: Attention-Enhanced Multi-Scale CNN Framework for Breast Cancer Classification in Histopathological Images

Breast cancer remains a leading cause of mortality among women worldwide, emphasizing the critical importance of early and accurate detection in improving patient outcomes and treatment plans. Histopathological analysis of breast tissue samples is the gold standard for breast cancer diagnosis, yet it is often labor-intensive, subjective, and prone to inter-observer variability. Recent advancements in deep learning have shown significant promise in enhancing the accuracy and efficiency of breast cancer classification. In this study, we introduce BreNet, a robust and empirically validated multi-scale CNN ensemble tailored for breast cancer classification in histopathological images. By combining three pre-trained CNN backbones with complementary channel and spatial attention mechanisms, BreNet refines feature representations across multiple scales. The model achieves strong and consistent performance on both the BreakHis and IDC datasets, reaching classification accuracies of 99.96% and 88.26% respectively, while maintaining low false-positive and false-negative rates. Through thorough evaluations including patient-level cross-validation, stain normalization, and magnification-specific analysis, BreNet demonstrates reliable generalization in clinically realistic settings. These results underscore its potential as a practical decision-support tool to assist pathologists in routine diagnostic workflows.