Equilibrium Optimization-Based Ensemble CNN Framework for Breast Cancer Multiclass Classification Using...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Breast cancer remains a leading cause of mortality in women.
• Early detection and accurate diagnosis are crucial for effective treatment and reduced mortality.
• Histopathological images are vital for breast cancer diagnosis and staging.

Purpose of the Study:

• To develop a high-performance deep learning model for diagnosing eight types of breast cancer.
• To address challenges in histopathological image analysis, including data imbalance and model overfitting.
• To propose a novel model, MultiHisNet, for improved breast cancer classification.

Main Methods:

• Utilized the BreakHis dataset of histopathological images.
• Fine-tuned 20 state-of-the-art deep learning models.
• Developed and evaluated 20 custom models, including the novel MultiHisNet.
• Created an ensemble model using best-performing models and an Equilibrium Optimizer.

Main Results:

• The novel MultiHisNet model achieved 94.69% accuracy in multi-class breast cancer classification.
• The proposed ensemble model reached 96.71% accuracy in eight-class breast cancer detection.
• The study identified key components for effective model performance, such as attention modules and residual links.

Conclusions:

• The developed deep learning models, particularly the ensemble approach, demonstrate high accuracy in breast cancer diagnosis.
• The findings suggest that the proposed model can serve as a valuable tool to assist pathologists.
• This research contributes to advancing automated breast cancer diagnosis through artificial intelligence.