The use of machine learning has become widespread in various areas related to medicine. Its application in MRI neuroimaging is in a state of permanent advance, and multiple initiatives point to a multimodal, multicenter implementation with a mainly translational focus. In this dissertation, we will discuss the main challenges and approaches to face them. We will analyze a framework to generate massive datasets, including the interaction with image acquisition systems (from a manual to an automated approach), the requirements at the clinical record level to develop translational methods, and harmonization to generate knowledge bases with images from different MRI machines. Then, we will present various techniques for the systematic extraction of robust features in a multimodal approach, and the use of classifiers, deepening into multiple classifier systems (MCSs), Random Forest (RF), and Deep Neural Networks (DNN).