Improved Breast Cancer Classification through Combining Transfer Learning and Attention Mechanism

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Pathology

Background:

• Breast cancer is a major global health concern, necessitating advanced diagnostic tools.
• Deep learning has shown promise in revolutionizing medical image analysis for disease detection.
• Accurate histopathological image classification is crucial for effective breast cancer diagnosis and treatment planning.

Purpose of the Study:

• To develop and evaluate a novel deep learning approach for breast cancer histopathological image classification.
• To enhance the interpretability and robustness of diagnostic models using attention mechanisms.
• To improve the accuracy of breast cancer detection in complex cases.

Main Methods:

• Utilized transfer learning with pre-trained deep convolutional neural network (CNN) models: Xception, VGG16, ResNet50, MobileNet, and DenseNet121.
• Integrated the convolutional block attention module (CBAM) to augment CNN models, focusing on localized features.
• Fine-tuned the models and evaluated their performance using accuracy, precision, recall, and F1 score on the BreakHis dataset.

Main Results:

• Attention mechanisms (AM) combined with the Xception model achieved test accuracies of 99.2% and 99.5%.
• The DenseNet121 model with AMs demonstrated a high test accuracy of 99.6%.
• The proposed methods outperformed previously studied approaches in breast cancer diagnosis.

Conclusions:

• The novel deep learning approach, incorporating attention mechanisms, significantly enhances breast cancer histopathological image classification accuracy.
• This method offers a robust and interpretable tool for computer-assisted pathological diagnosis.
• The findings suggest a promising advancement in automated breast cancer detection systems.