PSA-FL-CDM: A Novel Federated Learning-Based Consensus Model for Post-Stroke Assessment.

Area of Science:

• Artificial Intelligence in Healthcare
• Internet of Things (IoT) for Rehabilitation
• Data Privacy in Medical AI

Background:

• Internet of Things (IoT) devices generate vast datasets for smart rehabilitation.
• Patient privacy is critical in healthcare, especially with sensitive data.
• Existing AI models for stroke assessment may not adequately protect privacy.

Purpose of the Study:

• To propose a scalable AI model using federated learning for post-stroke assessment.
• To protect patient privacy during smart rehabilitation data analysis.
• To compare the performance of federated learning against centralized models.

Main Methods:

• Adoption of federated learning (PSA-FL-CDM) for scalable AI model development.
• Comparison with a centralized model (PSA-MNMF) on sensor- and camera-based datasets.
• Implementation of consensus models and eight clustering methods on each node.
• Utilization of the FedAvg algorithm for global model creation.

Main Results:

• The federated PSA-FL-CDM model significantly reduces computational time.
• Comparable performance is achieved with the federated model while preserving patient privacy.
• Federated learning enables collaborative model training without raw data sharing.
• Experiments conducted on both wearable and camera-based datasets.

Conclusions:

• Federated learning offers a scalable and privacy-preserving solution for AI-driven stroke assessment.
• The proposed PSA-FL-CDM model demonstrates efficiency and comparable accuracy.
• This approach facilitates collaborative healthcare analytics while upholding data confidentiality.