Our understanding of brain anatomy and physiology has advanced greatly thanks to the introduction of magnetic resonance imaging (MRI). As the technology develops and the amount of data multiplies, it is essential to develop methods to effectively and efficiently extract useful information from our data. However, scans are often collected under varying conditions, making analysis difficult. For this reason it is important to properly prepare the MRI for a quantitative and qualitative study. In this thesis, we investigate the use of deep learning models to prepare, process and analyse structural brain MRI scans. More specifically, we first introduce an unsupervised and interpretable method that register brain with high accuracy, especially in the context of large displacements. Then we show a segmentation strategy that requires only a single labeled example to train, while leveraging all the available unlabeled scans. Next, we present a novel method that warps brain template given a subject attributes. Finally, we discuss a scientific application of processed MRI and how our strategy can be useful to study neuroanatomical shape.