AMSD_VGGNet: A Lightweight Multi-Scale Attention Network for High-Accuracy Breast Cancer Histopathological Image Classification

Abstract

Early diagnosis of breast cancer is crucial for improving patient survival rates, but traditional pathological diagnosis heavily relies on subjective clinical experience, suffering from inefficiency and poor consistency. This paper proposes an improved VGG network model integrating multi-scale features and attention mechanisms for automated classification of breast cancer histopathological images. The model introduces a mixed-domain attention mechanism into the VGG16 backbone, enabling dynamic focus on critical pathological feature regions such as nuclear atypia. Simultaneously, it incorporates a dual-scale dilated convolution module to parallelly extract local details and global contextual information, enhancing multi-scale feature representation. Experimental results demonstrate that AMSD_VGGNet achieves classification accuracies of 99.71% on both BreakHis and ICIAR2018 datasets, with only 12.8% of VGG16's parameter count. Heatmap visualization indicates that its decision logic aligns closely with pathological standards. Furthermore, an interactive system interface developed using PySide6 framework supports high-resolution image loading and real-time classification response, providing an efficient and reliable intelligent auxiliary diagnostic tool for early breast cancer screening.