Federated Learning Empowered Real-Time Medical Data Processing Method for Smart Healthcare

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Machine Learning

Background:

• Computer-aided diagnosis (CAD) is crucial for smart healthcare, but deep learning faces challenges with chronologically ordered medical data.
• Traditional batch learning is resource-intensive for real-time data, while incremental learning suffers from catastrophic forgetting.

Purpose of the Study:

• To propose a real-time medical data processing method using federated learning to address limitations of traditional methods.
• To mitigate the catastrophic forgetting problem in incremental learning for medical AI applications.

Main Methods:

• A two-stage approach combining federated learning for model fusion (model stage) and exemplar-based knowledge review (exemplar stage).
• Federated learning integrates old and new models to prevent knowledge loss.
• Representative exemplars from old data aid new models in retaining past knowledge.

Main Results:

• The proposed method successfully learns a disease diagnosis model from a continuous medical real-time data stream.
• Model performance continuously improves as data volume increases.
• Catastrophic forgetting is effectively mitigated, outperforming traditional batch learning in resource efficiency.

Conclusions:

• The federated learning-based method offers an efficient solution for real-time medical data processing in CAD.
• This approach enhances diagnostic model accuracy and robustness while conserving computational resources.