A Hardware-Accelerated Federated Learning With Intuitionistic Fuzzy Security Measures for AI Adaptive Image Computing
Yechuan Lin, Lin Wang, Zhixuan Zhang, Xiaojian Liu · IEEE Access · 2025
This paper presents a hardware-accelerated federated learning framework with intuitionistic fuzzy security measures for adaptive medical image processing. The proposed system combines gaze shift path modeling with FPGA-optimized feature extraction to enable privacy-preserving analysis of distributed medical images. Using an enhanced BING objectness descriptor, the framework dynamically identifies diagnostically relevant regions while applying intuitionistic fuzzy measures to quantify uncertainty and protect sensitive data. The architecture implements a locality-preserved active learning strategy through parallelized fuzzy inference units on FPGA hardware, enabling real-time processing that maintains anatomical relationships during image adaptation. This approach achieves secure distributed computation without centralized data aggregation, making it suitable for clinical edge devices while complying with medical privacy regulations. The framework demonstrates three key advantages for medical image analysis: robust diagnostic region preservation through fuzzy-weighted feature importance, provable privacy guarantees via hesitation-based security measures, and efficient hardware acceleration for real-time performance. Experimental results show the system maintains high accuracy in clinical feature detection while significantly reducing vulnerability to membership inference attacks compared to conventional federated learning approaches. The FPGA implementation delivers low-latency processing suitable for educational and diagnostic applications, with measured performance meeting the power and speed requirements for deployment in distributed healthcare environments. By integrating intuitionistic fuzzy security with hardware-accelerated federated learning, this work advances the development of privacy-preserving medical imaging systems that can operate effectively across decentralized clinical networks.