Edge-Federated Quantum-Enhanced Generative Framework for Real-Time Arrhythmia Classification on ARM Architectures

Abstract

Accurate, low-latency, and privacy-preserving arrhythmia detection is essential to next-generation wearable and remote cardiac monitoring. In this study, we present Q-FGF, an Edge-Federated Quantum-Enhanced Generative Framework, for the real-time classification of ECG on low-power ARM platforms. The framework employs three novel componegurents in ECG classification: (1) a Quantum-Inspired Variational GAN (Q-VGAN) for the synthesis of morphologically consistent minority class ECG signals; (2) a secure federated learning protocol, which uses additive homomorphic masking and efficient aggregation, to facilitate distributed training in a privacy-preserving manner; and (3) a quantum-hybrid Deep Belief Network (Q-DBN) utilizing parameterized nonlinear encoding and 8-bit quantization to perform ultra-efficient inference. Key to Q-VGAN's generation of morpho-temporal fidelity is its hybrid feature mapping and supervised sequence loss approach. The federated pipeline incorporates local, encrypted computation while supporting compliance with GDPR and HIPAA. The optimized Q-DBN achieves real-time inference (<95 ms) and low energy (<200 mW) on an STM32F4 microcontroller. Evaluation on three datasets (MIT-BIH, PTB-XL, CPSC 2018) under centralized and federated settings establishes Q-FGF as a practical solution for privacy-preserving, clinically deployable ECG monitoring at the edge, showcasing up to 18% greater minority class recall rates and 12% higher macro-F1 scores than state-of-the-art (DCGAN+DBN, TimeGAN, MobileNetV3) baselines.