Deep-Reinforcement-Learning-Driven Patient State Analysis and Resource Management in Near-Field IoE Healthcare Networks

Abstract

Integrating Internet of Everything (IoE) technologies with near-field (NF) communication has revolutionized healthcare monitoring systems, offering unprecedented opportunities for precise patient state tracking and resource optimization. However, managing healthcare IoE resources efficiently while maintaining high-quality patient monitoring in NF environments presents significant challenges, particularly in multipatient scenarios where traditional approaches struggle with continuous control requirements and high-dimensional decision spaces. This article proposes probabilistic patient modeling with twin delayed deep deterministic policy gradient using Gaussian Wasserstein distance (PPM-TD3-GWD), a novel deep-reinforcement-learning framework for healthcare resource management in NF-enabled IoE networks. Our approach combines probabilistic patient state modeling with advanced reinforcement learning techniques, leveraging NF communication characteristics to optimize monitoring accuracy and resource efficiency. The framework employs Gaussian Wasserstein distance to measure probabilistic differences between prior and posterior patient state distributions, enabling precise quantification of information gain from sensor measurements. Gaussian Wasserstein distance computation guides sensor positioning to maximize measurement informativeness while minimizing resource consumption by incorporating spherical wavefront characteristics of NF communication. Experiments in single-patient and multipatient scenarios, including detailed diabetes patient state analysis, PPM-TD3-GWD demonstrates concrete improvements: tracking error decreased from 0.5091 mg/dL to 0.3842 mg/dL (24.53% reduction), resource efficiency improved from 72.8% to 85.4% (17.40% enhancement), and critical glucose event detection rate reached 95% compared to 70% in baseline approaches.