Biofeedback-augmented LLM framework for personalized dietary risk assessment

This is the authors' abstract. We don't add key points for this paper.

We propose B-QoS-DietNet, a QoS-aware, WBAN-enabled intelligent dietary risk assessment framework for life-critical healthcare applications. The B-QoS-DietNet model effectively tackles critical issues present in the current healthcare-monitoring systems using wearable devices. This is achieved by tackling physiological dynamism, resource constraints of WBANs, limited real-time adaptability of current systems, and the absence of closed-loop intelligence in current AI-based models. The problems above affect QoS attributes such as responsiveness, reliability, adaptiveness, and personalization. The novel approach involves combining real-time physiological sensing from WBAN nodes with a biofeedback-enabled large language model. Using this framework, dietary risk assessment can be carried out in a more efficient and accurate manner by considering patients’ medical histories and physiological characteristics. The information from sensors including glucose levels, blood pressure, and heart rates is integrated in an edge-centric approach where diet risk is recalibrated based on semantic analysis, behavioral analysis, and multi-objective optimization techniques. Furthermore, the use of biofeedback solves cross-disease conflict and achieves user-personalized recommendations. Unlike the current static large language model and rule-driven systems, B-QoS-DietNet offers adaptive response to dietary risk assessment in a personalized manner under dynamic health situations. Through simulations on WBAN scenarios, the model offers significantly better results as illustrated in the experiment section where we achieve up to 7.8% improvement in personalization gain and classification accuracy.