A Scalable Federated Learning Approach for Collaborative Smart Healthcare Systems With Intermittent Clients Using...

Area of Science:

• Computer Science, Artificial Intelligence, Machine Learning, Healthcare Informatics, Data Privacy

Background:

• The healthcare industry faces significant cybercrime and privacy risks due to sensitive, distributed health data.
• Existing data protection methods struggle with the accuracy and sustainability required for modern healthcare.
• Decentralized healthcare systems are hindered by intermittent remote clients and imbalanced datasets.

Purpose of the Study:

• To implement a scalable federated learning (FL) framework for interactive smart healthcare systems.
• To address challenges posed by intermittent clients and imbalanced datasets in remote healthcare settings.
• To enhance patient diagnostic model development through collaborative, privacy-preserving data utilization.

Main Methods:

• A federated learning (FL) framework was implemented for smart healthcare systems using chest X-ray images.
• Data augmentation techniques were employed to balance imbalanced datasets for local model training.
• The framework was tested with varying numbers of clients (5-18) and data sizes to assess performance under intermittent connectivity.

Main Results:

• The proposed FL approach demonstrated competitive performance in handling both intermittent clients and imbalanced datasets.
• Experimental results validated the framework's scalability and effectiveness across diverse testing scenarios.
• The FL method successfully balanced datasets and managed client dropouts/joins during training.

Conclusions:

• Federated learning offers a viable, privacy-preserving solution for developing robust diagnostic models in decentralized healthcare.
• The study encourages medical institutions to collaborate using private data for advanced patient diagnostics.
• This FL framework supports the development of powerful diagnostic tools by overcoming data limitations in remote healthcare.