Mathematical symbols and words vary so much, it can be quite difficult to read a mathematical expression or equation from an image. In this instance, reading an equation in mathematics is understood to be the process of composing a textual explanation of the equation. We present the MED model, a creative complete trainable deep neural network-based approach that learns to produce a textual description for mathematical equation images. MED was created by the natural picture captioning challenge in computer vision. Two neural networks are implemented in our model: a RNN with an focus mechanism that creates descriptions related to the input mathematical equation images, and a CNN that works as an encoder, extracting features from the input mathematical expression images.