Numerical endocrinology: Portable solutions of cortisol on small edge devices using AI

The time varying concentration of blood cortisol depends predominantly on secretion, clearance, and reversible binding to plasma proteins as well as diffusion between the blood volume and extravascular volume which comprises an additional compartment. In the blood it appears free and adsorbed to two different blood proteins, albumin and cortisol binding globulin (CBG) regarded as three compartments. The concentrations of cortisol in the total of the four compartments are related by a mathematical model consisting of four equations. In common use in endocrinology the term "compartment" refers to the state rather than a physical region. One such state is cortisol bound to albumin in the blood. The objective here is to examine these concentrations for a variety of subjects using numerical methods and to develop a fast, efficient Artificial Intelligence (AI) method of obtaining predictions of the cortisol system on small, portable edge devices. The mathematical model is represented by four simultaneous, non-linear differential equations, the four-compartment model. Performance has been confirmed using human subject data for impulse inputs at time zero with suppressed cortisol secretion. The suppression takes one variable out of the mathematical model under investigation. Supervised learning of an ANN was developed here to replace previous large, slow, and inefficient numerical methods. The ANN provided excellent predictive accuracy well within the range of measurement accuracy. Parameter predictions included (i) free cortisol elimination rate constant α (s-1), (ii) a diffusion and barrier geometry constant k (L/s), and (iii) effective extravascular volume fraction (Vf). The ANN can be deployed on a small edge device for easy use in clinical settings.