Efficient Breast Cancer Classification Network with Dual Squeeze and Excitation in Histopathological Images

Area of Science:

• Digital pathology
• Computational oncology
• Artificial intelligence in medicine

Background:

• Current medical imaging techniques like mammography, ultrasound, and MRI lack cellular-level detail for comprehensive cancer microenvironment analysis.
• This limitation hinders accurate breast cancer subtype classification, necessitating advanced analytical approaches.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN)-based method for accurate breast cancer classification using hematoxylin and eosin (H&E) whole slide images (WSIs).
• To classify breast cancer into binary (benign/malignant) and eight distinct subtypes based on histopathology.

Main Methods:

• A novel CNN architecture incorporating fused mobile inverted bottleneck convolutions (FMB-Conv) and mobile inverted bottleneck convolutions (MBConv) with a dual squeeze and excitation (DSE) network was employed.
• A pre-trained EfficientNetV2 network served as the backbone, featuring a modified DSE block to enhance feature extraction by integrating spatial and channel-wise attention.
• The method was trained and validated on the BreakHis dataset, evaluating performance across multiple magnification levels.

Main Results:

• The proposed CNN method demonstrated superior performance compared to ResNet101, InceptionResNetV2, and standard EfficientNetV2.
• Achieved higher precision, recall, and F1-score for both binary and multi-class breast cancer classification tasks.
• Effectiveness was validated across various magnification levels within the histopathology images.

Conclusions:

• The developed CNN approach offers a robust and accurate method for breast cancer classification from H&E stained WSIs.
• This technique has the potential to significantly improve the understanding and classification of breast cancer subtypes in digital pathology.
• The findings highlight the efficacy of integrating advanced CNN architectures with attention mechanisms for complex histopathological image analysis.