A Deep Learning Method for Breast Cancer Classification in the Pathology Images

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Oncology

Background:

• Breast cancer is a leading global health concern for women.
• Deep learning shows promise for early breast cancer diagnosis.
• Common Convolutional Neural Networks (CNNs) often overfit breast pathology datasets.

Purpose of the Study:

• To propose a novel framework, AlexNet-BC, to mitigate overfitting in breast cancer classification.
• To enhance classification accuracy using deep learning methodologies.
• To improve the robustness and generalization of diagnostic models.

Main Methods:

• Developed the AlexNet-BC model, pre-trained on ImageNet and fine-tuned on augmented breast pathology data.
• Devised an improved cross-entropy loss function to penalize overconfident predictions.
• Validated the model on BreaKHis, IDC, and UCSB datasets across various magnifications.

Main Results:

• The AlexNet-BC model demonstrated superior performance compared to state-of-the-art methods.
• The proposed approach effectively reduced overfitting in breast pathology image classification.
• Achieved strong robustness and generalization capabilities on multiple datasets.

Conclusions:

• The AlexNet-BC model offers a promising solution for accurate breast cancer classification.
• The improved loss function enhances prediction reliability for clinical applications.
• This framework is suitable for histopathology clinical computer-aided diagnosis systems.