Optimized federated learning framework with RegNetZ and Swin-Transformer for multimodal pancreatic cancer detection1

Area of Science:

• Artificial Intelligence in Oncology
• Medical Imaging Analysis
• Federated Learning for Healthcare

Background:

• Pancreatic cancer has a low survival rate due to late diagnosis and challenges in early detection using CT/MRI.
• Centralized deep learning in healthcare faces significant privacy and data-sharing obstacles.
• Early detection is crucial for improving patient outcomes in pancreatic cancer management.

Purpose of the Study:

• To develop a federated learning framework for automated pancreatic cancer detection, subtype classification, and prognosis prediction.
• To integrate multimodal data (CT, MRI, histology, genomics, clinical records) for comprehensive analysis.
• To address privacy concerns by enabling decentralized model training without raw data exchange.

Main Methods:

• A federated learning framework combining RegNetZ and Swin-Transformer for feature extraction and dependency modeling.
• Hybrid Aquila-Grey Wolf Optimizer (HA-GWO) for efficient hyperparameter tuning.
• Evaluation across 5-7 simulated client institutions using multimodal patient data.

Main Results:

• Achieved high performance metrics: 99.2% accuracy, 98.9% sensitivity, 99.0% precision, and 99.4% AUC.
• Outperformed CNN-only and transformer-only baselines in detection, classification, and prognosis prediction.
• Demonstrated significant reduction in false positives and negatives for improved diagnostic reliability.

Conclusions:

• The proposed federated learning framework enhances pancreatic cancer diagnosis accuracy while preserving patient privacy.
• The system offers a scalable and cost-effective solution for real-time detection and precision oncology in federated environments.
• Optimized hyperparameters (learning rate 0.003, batch size 64) ensure efficient model performance.