An apparatus for monitoring blood glucose comprising an invasive component for invasively measuring blood glucose and a non-invasive component, including color image sensor(s) to generate images from absorption of light that traversed the tissue, to receive a body part and generate a non-invasive blood glucose reading. Processor(s) convert the images into a vector V associated with a particular at least one invasive blood glucose measurement gk1, form a regular learning matrix, λ, implement a noninvasive isolation mechanism of the tissue glucose level by unique association of the vector Vk with an invasive blood glucose level, determine a neural network from the learning set λ by pairing vectors into a branch and forming multiple branches into loops, wherein two vectors are paired if they have a pre-defined similarity in the blood glucose levels that each are associated with; and calibrate the neural network by having it pass at least one test.