Federated Learning Architecture for 3D Breast Cancer Image Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Breast cancer diagnosis relies heavily on mammography, but interpretation can be challenging.
• Automated detection methods are crucial for improving diagnostic accuracy and efficiency.
• Early detection significantly enhances patient survival rates for breast cancer.

Purpose of the Study:

• To develop and evaluate an advanced automated breast cancer detection system.
• To integrate 3D mammographic imaging with Federated Learning (FL) for privacy-preserving, decentralized model training.
• To compare the performance of Convolutional Neural Networks (CNNs), Transfer Learning models, and AutoEncoders for this task.

Main Methods:

• Utilized 3D mammographic imaging data for model training and evaluation.
• Implemented and compared various machine learning models: CNNs, Transfer Learning (VGG16, VGG19, ResNet50), and AutoEncoders (AEs).
• Employed Federated Learning (FL) to enable decentralized and privacy-preserving model training across multiple institutions.

Main Results:

• The CNN model achieved a high accuracy of 97.30%.
• Combining the CNN with Federated Learning (CNN-FL) slightly improved accuracy to 97.37%, demonstrating robust predictive performance.
• Transfer Learning models and AutoEncoders showed lower accuracies, ranging from 48.83% to 89.24%.

Conclusions:

• The CNN-FL framework is a highly effective tool for automated breast cancer detection.
• This approach successfully balances high diagnostic accuracy with crucial data security.
• The findings highlight the potential of federated learning in enhancing medical imaging analysis while maintaining patient privacy.