Federated learning with swarm intelligence for efficient and secure medical image analysis

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Federated Learning

Background:

• Healthcare collaborative learning is hindered by strict regulations and fragmented data.
• Analyzing medical images requires robust, privacy-preserving methods.
• Existing frameworks struggle to balance privacy, communication efficiency, and diagnostic accuracy.

Purpose of the Study:

• To develop a federated learning framework augmented with swarm intelligence for enhanced medical image analysis.
• To optimize hyperparameters, feature selection, and client aggregation weights simultaneously.
• To ensure a balance between patient privacy, communication costs, and classification accuracy.

Main Methods:

• Combined Particle Swarm Optimization (PSO) and Firefly Algorithm (FA) with deep Convolutional Neural Networks (CNNs).
• Tested the framework on COVID-19, monkeypox, and breast cancer datasets across simulated healthcare institutions.
• Implemented robust privacy measures and statistical validation.

Main Results:

• Achieved high diagnostic accuracy: 96.71% for COVID-19, 96.06% for monkeypox, and 97.0% for breast cancer.
• Reduced communication rounds by 25-30% and demonstrated resilience against noise and attacks.
• Privacy-utility analysis showed acceptable trade-offs with accuracy above 94%.

Conclusions:

• The proposed federated learning framework effectively enhances medical image analysis while preserving patient privacy.
• This approach is suitable for smaller healthcare settings, enabling AI adoption without compromising data security.
• The study demonstrates a viable solution for secure and efficient collaborative medical AI.