Dynamic-Fusion-Based Federated Learning for COVID-19 Detection.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Machine Learning

Background:

• Machine learning-based medical image analysis offers efficient COVID-19 detection.
• Patient privacy concerns restrict data sharing, leading to insufficient training datasets.
• Federated learning enables collaborative model training without data exchange.

Purpose of the Study:

• To propose a novel dynamic fusion-based federated learning approach for COVID-19 detection using medical images.
• To enhance communication efficiency and model performance in federated learning systems for medical image analysis.

Main Methods:

• Designed a dynamic fusion-based federated learning system architecture for medical image analysis.
• Developed a dynamic fusion method to select participating clients based on local model performance and training time.
• Compiled a dataset of medical diagnostic images for COVID-19 detection.

Main Results:

• The proposed dynamic fusion approach demonstrated feasibility.
• Achieved superior performance compared to default federated learning settings.
• Showcased improvements in model performance, communication efficiency, and fault tolerance.

Conclusions:

• The dynamic fusion-based federated learning approach is effective for COVID-19 detection from medical images.
• This method addresses data heterogeneity and communication costs in federated learning.
• It offers a promising solution for privacy-preserving medical image analysis.