Neural network-based prediction and optimization of estradiol release from ethylene-vinyl acetate membranes

Drug-delivery systems, with predictable delivery rates, were designed using an artificial neural network-based optimization algorithm. A two-chamber diffusion cell was used to study the permeation of estradiol through ethylene-vinyl acetate copolymer membrane. The explanatory variables were the vinyl acetate (VA) content of the membrane, poly(ethylene glycol) (PEG) - solvent-composition, and membrane thickness. After deriving a neural network model to predict estradiol delivery rates as a function of these input variables, a constrained optimization procedure was applied to estimate the membrane/vehicle properties necessary to achieve a prescribed dosage. The results compared adequately well with experimental data with 71% of the data agreeing within one standard deviation. Input sensitivity analysis showed that at specific VA levels, drug delivery was more sensitive to changes in PEG compositions. The non-uniqueness of the inversion method and the accuracy of the procedure were investigated using neural network-based two-dimensional contour plots. The methodology proposed could be used to design customized polymer-based drug-delivery systems that meet specific end-user requirements.