Analyzing Energy Requirements of Meta-Differential Evolution for Future Wearable Medical Devices

Recent advances in clinical engineering include development of physiological models to deliver optimized healthcare. Physiological model comprises a number of equations to relate biomedical signals. Each equation contains a set of coefficients. Determining the coefficients is a complex task as the models are non-linear. Therefore, development of the models must be accompanied by a development of methods to determine model coefficients. With the advent of wearable medical devices, we have to consider energy requirements of the models and the methods. Considering an illustrative case of type-1 diabetes mellitus patients, we already demonstrated that Meta-Differential Evolution outperforms analytical methods, when determining coefficients of glucose dynamics. In this paper, we analyze convergence of the Meta-Differential Evolution, running time and associated power consumption on a single board computer with a system-on-a-chip—Cortex-A8 AM335x. Based on the analysis, we recommend splitting the process of determining the coefficients into two phases. First phase determines the initial, per-patient optimized coefficients. Second phase is an energetically efficient update of these coefficients with new, continuously measured signal of the patient. Meta-Differential Evolution searches for optimal coefficients by evolving a number of generations of candidate coefficients, using a number of evolutionary strategies. We demonstrate that the proposed approach significantly reduces the number of candidate coefficients to evaluate, while achieving the desired accuracy. This positively reflects in the lifetime of wearable device’s battery. Specifically, calculating coefficient’s update took 0.05 Ws only. It shows the feasibility of using Meta-Differential Evolution with its improved accuracy for blood glucose calculations in a wearable device.