A major issue in the association of genes to imaging phenotypes is the high dimension of both the genetic data and imaging data. In this talk, we discuss our recent work addressing the latter problem. A key concept of our work is a neuroimaging genetic pipeline where we separate the neuroimaging genetic association in three distinct steps: first is image processing, then, neuroimaging feature extraction and finally the genetic association study. The novelty of our approach is using a neural network classifier in order to accomplish the second step; extracting neuroimaging features that are related with AD. One could think of our model as an AutoEncoder where the decoder is replaced by a prediction function. To conclude, we compared the predictive power of the features automatically extracted by our approach to expert-selected features.