In the recent years, deep learning based on convolutional neural networks have outperformed the state of the art in several image processing tasks, in particular for the image classification. In order to well train the network and hence, have a successful deep neural network (DNN), one need thousands of labeled images. However, in biomedical imaging it is often impossible to access to have a such huge database. In a such case, data augmentation is the technique to artificially create more labeled images. The goal of projet is to design and to implement a data augmentation module and to train the neural network with desired invariance and robustness properties, when only few training images are available. The objective of this project is to study the performance of the DNN while using data augmentation.