Enhancing Breast Cancer Detection with HNet: A Hybrid Deep Learning Framework

Abstract

Breast cancer remains one of the leading causes of cancer-related mortality among women globally, making early and accurate diagnosis critical for effective treatment. While traditional convolutional neural networks (CNNs) are proficient in extracting local texture features, they often struggle to capture global contextual dependencies and spatial hierarchies inherent in histopathological images. To overcome these limitations, we propose HNet, a novel hybrid deep learning architecture designed to leverage the complementary strengths of multiple techniques. HNet combines EfficientNet for scalable and efficient local feature extraction, Advanced Vision Transformers (AVT) for global context modeling, and Capsule Networks for relational reasoning and spatial hierarchy preservation. This fusion of architectures aims to enhance diagnostic performance and improve interpretability. Evaluated on the BreakHis dataset across multiple image resolutions and data split configurations, HNet demonstrated an accuracy up to 97.52%, showcasing enhanced classification accuracy and generalization. Ablation studies further validated the contribution of each module, highlighting the potential of hybrid deep learning frameworks in enabling robust, real-world breast cancer diagnosis.