Energy-efficient early emergency detection for healthcare monitoring on WBAN platform

This dissertation introduces an innovative structure aimed at improving anomaly detection and predictive analyses in Wireless Body Area Networks (WBANs), a crucial technology within the realm of digital healthcare. Motivated by the need to improve diagnostic precision and clinical decision-making, especially in environments constrained by the computational limitations of edge devices, this research aims to revolutionise patient monitoring systems. The research begins with a comprehensive review of current WBAN technologies and their applications in healthcare. It identifies a distinct gap in the ability of these systems to adapt to the dynamic and complex nature of patient health monitoring. Traditional WBAN methodologies, heavily reliant on static thresholds and centralised cloud-based processing, often fall short of effectively managing the nuanced and varied data derived from patient monitoring, leading to real-time responsiveness and energy efficiency challenges.

The research progresses from static to dynamic threshold to address these challenges, enhancing the system's adaptability to fluctuating health indicators. The Multi-Level Classification Threshold Algorithm (MLCTA) was formulated to refine the classification of health-related data. The study subsequently presents a compound method that combines threshold-based techniques with linear regression analysis. This integration significantly bolsters the model's predictive capacity for health incidents by providing a more profound comprehension of vital sign patterns. When used in conjunction with actual patient data, this approach notably heightens the precision of health event forecasts.

The framework includes a series of progressively advanced algorithms: The Modified Adaptive Local Emergency Detection (MALED) lays the groundwork with its adaptive response to health data changes. This is enhanced by the Differential Change Analysis (DCA), which introduces sensitivity to the rate of change in vital signs for early anomaly detection. The Local Emergency Detection Algorithm Using Adaptive Sampling (LEDAS) further optimises this framework by implementing adaptive sampling based on the patient's health status, ensuring efficient data collection. The pinnacle of this progression is the Sequential Multi-Dimensional Trend Analysis (SMDTA), which offers a comprehensive multi-dimensional analysis of health data, identifying intricate patterns and relationships among various vital signs for precise health predictions. Additionally, incorporating dynamic thresholds across these algorithms refines anomaly detection, making the system more flexible and responsive to changing patient health dynamics. Together, these algorithms represent a significant leap from basic monitoring systems to advanced networks capable of sophisticated multi-dimensional health analysis.
Empirical evaluation using actual patient data from clinical databases demonstrated the superior efficacy of the proposed framework. Notably, the hybrid approach combining linear regression with threshold-based methods achieved near 96% accuracy in anomaly detection, significantly reducing the false-positive rate to 2%. Furthermore, the optimised local emergency detection strategies led to an average 85% reduction in data transmissions, contributing to a 19% decrease in energy consumption compared to existing methods, thereby underscoring the system's suitability for energy-constrained environments. The results of this research highlight not only the potential of advanced WBAN systems in enhancing healthcare delivery but also pave the way for future developments in medical technology. The proposed framework and its algorithms open new avenues in clinical decision-making, offering robust, efficient, and user-friendly solutions for healthcare professionals and patients.