DRDA-Net

DRDA-Net classifies breast cancer from histopathological images using a dense residual dual-shuffle attention deep learning architecture to improve feature discrimination.

Key Features:

• Dual-shuffle Attention Mechanism: Incorporates a dual-shuffle attention-guided approach inspired by the ShuffleNet bottleneck unit to enhance focus on relevant features within histopathological images.
• Channel Attention Mechanism: Includes a channel attention mechanism to prioritize informative channels during feature extraction and refine detection of complex tissue patterns.
• Densely Connected Blocks: Employs densely connected blocks to mitigate overfitting and the vanishing gradient problem by facilitating efficient information flow for robust learning on relatively small datasets.

Scientific Applications:

• Breast cancer classification across magnifications: Evaluated on the BreaKHis dataset with reported accuracies of 95.72% at 40x, 94.41% at 1000x, 97.43% at 200x, and 98.1% at 400x magnification for histopathological image classification.

Methodology:

Integration of dual-shuffle and channel attention mechanisms with densely connected blocks within a deep learning framework to process and analyze histopathological images.