Multi-Objective NSGA-II Optimization of Latency, Accuracy, and Energy in Edge-Computing-Enabled 5G Networks for Remote Cardiac Prognostics

Abstract

Remote cardiac prognostics increasingly relies on continuous physiological sensing coupled with real time inference to detect life-threatening arrhythmias before clinical deterioration occurs. Cloud-centric deep learning pipelines deliver high diagnostic accuracy but incur end-to-end latencies that are unacceptable for Ultra-Reliable Low-Latency Communication (URLLC) class cardiac alerting, while resource-constrained edge devices sacrifice accuracy for speed. This paper proposes an edge computing-enabled 5G architecture that jointly optimizes inference latency and diagnostic accuracy for remote cardiac prognostics using a constrained multi-objective formulation solved through a Non dominated Sorting Genetic Algorithm II (NSGA-II) based model-selection and offloading controller. A hybrid CNN–LSTM prognostic model is compressed through structured pruning and INT8 quantization and dynamically deployed across a three-tier patient–edge–cloud pipeline, with a confidence-threshold policy governing selective cloud offloading over a network-sliced 5G URLLC bearer. Simulation-based evaluation on ECG beat-classification workloads shows that the proposed scheme attains 95.7% diagnostic accuracy at 42 ms end-to-end latency, a 57.1% latency reduction relative to cloud-only inference (98 ms) with only a 2.2 percentage-point accuracy loss, and sustains latency below the 50 ms URLLC target for up to 150 concurrently monitored patients per edge cell. The results indicate that adaptive, confidence-aware edge–5G orchestration offers a practical latency accuracy operating point for scalable, real-time cardiac telemonitoring.