Related Experiment Video
Updated: May 6, 2026
Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities
Published on: October 27, 2023
Adaptive homomorphic federated learning framework for multi-institutional medical imaging with optimized diagnostic
L Josephine Usha1, K R Saranya2, Y Suganya2
1Department of Computer Science and Engineering, School of Computing, SRM Institute of Science and Technology, Tiruchirappalli, Tamil Nadu, India. josephineusha@gmail.com.
The Next-Generation Adaptive Secure Federated Learning (NASFL) framework enhances medical AI diagnostics by securely fusing multi-modal imaging data. This AI approach achieves high accuracy and robustness, overcoming limitations of traditional federated learning for clinical deployment.
Area of Science:
- Medical Artificial Intelligence
- Federated Learning
- Medical Imaging Analysis
Background:
- Growing use of AI in medicine necessitates privacy-preserving diagnostic systems.
- Traditional federated learning (FL) faces challenges with heterogeneous data, noise, and communication overhead, limiting clinical use.
- Existing FL solutions struggle with multi-modal data integration and robustness in real-world clinical settings.
Purpose of the Study:
- Introduce the Next-Generation Adaptive Secure Federated Learning (NASFL) framework for ultra-accurate, scalable, and secure multi-institutional medical AI.
- Address limitations of traditional FL in handling diverse medical data and ensuring patient privacy.
- Develop a clinically deployable AI platform for robust medical diagnostics.
Main Methods:
- Employed multi-level homomorphic encryption (MLHE) and stochastic differential privacy for patient confidentiality.
- Utilized a transformer-guided ResNet backbone for adaptive multi-modal feature fusion (X-ray and CT data).
- Implemented institution-specific focus, trust-based aggregation, top-k gradient compression, and adaptive learning rates for efficiency and robustness.
Main Results:
- Achieved 99.6% diagnostic accuracy on multi-institutional datasets (NIH Chest X-ray14, LIDC-IDRI CT).
- Demonstrated low convergence time (approx. 65 communication rounds) and robustness in heterogeneous settings.
- Validated strong privacy assurance, surpassing traditional FL and state-of-the-art privacy-preserving methods.
Conclusions:
- NASFL establishes a clinically sound platform for secure multi-institutional medical AI.
- The framework sets a new benchmark for scalable, high-accuracy, and robust federated medical diagnostics.
- NASFL enables wide-scale deployment of advanced AI in healthcare settings.