Weighted aggregation through probability based ranking: An optimized federated learning architecture to classify...

Area of Science:

• Artificial Intelligence in Medicine
• Medical Diagnostics
• Health Informatics

Background:

• Respiratory diseases are a leading global health concern, often diagnosed via subjective auscultation.
• Current diagnostic methods require significant expertise and are prone to misinterpretation.
• Medical data privacy concerns limit the development of AI diagnostic tools.

Purpose of the Study:

• To develop an AI-based system for accurate respiratory disease classification using lung sounds.
• To implement a privacy-preserving Federated Learning (FL) approach for distributed medical data.
• To enhance FL model accuracy and efficiency through a novel aggregation algorithm.

Main Methods:

• Utilized Generative Adversarial Networks (GANs) for data synthesis and privacy preservation.
• Employed deep learning models trained on spectrograms of synthesized lung sounds.
• Introduced a novel Weighted Aggregation through Probability-based Ranking (FedWAPR) algorithm for FL optimization.

Main Results:

• Achieved a test accuracy of approximately 92% in classifying respiratory diseases and heart failure.
• Demonstrated superior performance of the FedWAPR algorithm over FedAVG in FL aggregation.
• Validated the model's effectiveness in a decentralized training environment, matching centralized model accuracy.

Conclusions:

• The AI-driven approach enables accurate, privacy-preserving respiratory disease detection from lung sounds.
• FedWAPR significantly improves federated learning aggregation for medical applications.
• This technology offers a secure and efficient alternative for respiratory health monitoring.