FoSoHA-Net: Deep Learning for Accurate Breast Cancer Classification Leveraging Hybrid Attention Mechanisms
Daniel Addo, Mugahed A. Al–antari, Shijie Zhou, Eric Ashalley, Gladys Wavinya Muoka, Obed Tettey Nartey · 2024
Breast cancer diagnosis is crucial for timely treatment and management. This study introduces FoSoHA-Net, a novel methodology for precise breast cancer classification using histopathological images. FoSoHA-Net integrates a hybrid attention module (HAM) and depthwise separable convolution to enhance feature extraction and improve classification accuracy. The hybrid attention module combines first-order pooling attention (FoPA) and second-order pooling attention (SoPA) to leverage global and local image information, augmenting the model’s discriminative power. Through comprehensive evaluation of the BreaKHis dataset, FoSoHA-Net demonstrates superior performance in accurately classifying histopathological images across various magnification levels. Comparative analysis against established pre-trained models further validates the effectiveness of FoSoHA-Net in breast cancer classification tasks. Overall, FoSoHA-Net shows promise in advancing breast cancer diagnosis through innovative AI-driven approaches.