The advanced treatment of the type I diabetes mellitus through closed-loop systems, such as artificial pancreas is one of today's greatest challenges. In this work the authors have tried to develop a predictive method which can be incorporated in an optimal system for blood glucose-insulin regulation. Using the neural network approach it is possible to compose a prediction model, very useful to control an insulin pump. Also, the artificial neural networks can be used to test and validate the medical decisions. The results of this study can be applied also, to other physiological systems, as it offers important data for the medical practice and it contributes to introduce the computer assisted diagnosis like a current medical method.