Abstract
The rapid digitalization of healthcare demands intelligent, low-latency, and privacy-preserving systems capable of operating at the network edge. This study introduces a unified Edge-AI framework that seamlessly combines dual wireless connectivity (LoRaWAN + 5G), federated learning (FL), Proof-of-Authority (PoA) block chain, and homomorphic encryption (HE) to achieve secure real-time anomaly detection in patient monitoring. A quantized CNN-LSTM model was deployed on NVIDIA Jetson Nano devices and trained using a synthetic dataset statistically modelled from the MIT-BIH Arrhythmia Database, capturing vital signals such as heart rate, temperature, and oxygen saturation. The integrated system attained 91.9% accuracy and 90.8% F1-score, with only an 8.7% latency overhead attributed to HE operations. A paired two-tailed t-test (p < 0.01) confirmed that these gains are both statistically and clinically significant, indicating reliable diagnostic performance under constrained conditions. Beyond performance, the proposed framework ensures end-to-end data confidentiality, tamper-proof auditability, and energy-efficient edge inference, offering a scalable pathway toward trustworthy, next-generation smart-healthcare ecosystems.