Adaptive federated learning for personalised affective pain management

Abstract

Conventional pain management strategies, particularly those relying on prolonged pharmaceutical interventions, are often burdened by adverse side effects and the development of tolerance. This study introduces an innovative, non-pharmacological framework integrating advanced machine learning techniques with immersive virtual reality (VR) distraction therapy. Central to our approach is a closed-loop, data-driven system that leverages affective computing to continuously monitor physiological indicators— such as heart rate variability (HRV) and beats per minute (BPM)—using non-intrusive wearable sensors. These real-time measurements inform a compact neural network deployed on each device, which is dynamically updated via a federated learning scheme. This distributed learning paradigm allows the global model to aggregate local updates while preserving patient privacy, thereby enhancing the personalised adaptation of the VR environment to the patient’s emotional state.The adaptive system is designed to maintain the patient within an optimal ‘flow’ state, effectively distracting from pain signals by tailoring the immersive experience in real time. In a controlled experiment employing a modified sphygmomanometer test, our approach demonstrated a 23% improvement in pain tolerance (p<0.05) over standard VR therapy and a 50% improvement over traditional care. The analysis confirms that incorporating federated learning yields a further enhancement in performance, underlining the potential of continuous model retraining to overcome the limitations of static, generic models. Our findings suggest that the proposed framework offers a safe and effective adjunct to conventional pain management and sets a new benchmark for personalised therapeutic interventions driven by advanced machine learning and data science methodologies.