Tri-branch feature pyramid network based on federated particle swarm optimization for polyp segmentation

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Computer Science

Background:

• Deep learning shows promise in medical imaging but is limited by small datasets due to privacy concerns.
• Legal and ethical constraints restrict access to medical data, hindering deep learning applications.
• Effective utilization of decentralized medical data is crucial for advancing AI in healthcare.

Purpose of the Study:

• To develop a privacy-preserving federated learning approach for medical image processing.
• To enhance the efficiency of decentralized data utilization in federated learning.
• To improve the stability and convergence speed of deep learning models in medical imaging.

Main Methods:

• Proposed Federated Particle Swarm Optimization (FPSO) algorithm for efficient, private model training.
• Introduced Tri-branch feature pyramid network (TFPNet) for stable federated learning and fast convergence.
• Conducted experiments on four public datasets: CVC-ClinicDB, Kvasir, CVC-ColonDB, and ETIS-LaribPolypDB.

Main Results:

• FPSO outperformed single-dataset training and Federated Averaging on independent, scattered data.
• TFPNet demonstrated faster convergence compared to other models.
• TFPNet achieved superior segmentation accuracy in medical image analysis.

Conclusions:

• The proposed FPSO algorithm effectively addresses privacy concerns and enhances decentralized data utilization.
• TFPNet provides a stable and efficient architecture for federated learning in medical imaging.
• This approach significantly improves deep learning model performance for medical image segmentation.