Recent advances in deep learning methods have come to define the\nstate-of-the-art for many medical imaging applications, surpassing even human\njudgment in several tasks. Those models, however, when trained to reduce the\nempirical risk on a single domain, fail to generalize when applied to other\ndomains, a very common scenario in medical imaging due to the variability of\nimages and anatomical structures, even across the same imaging modality. In\nthis work, we extend the method of unsupervised domain adaptation using\nself-ensembling for the semantic segmentation task and explore multiple facets\nof the method on a small and realistic publicly-available magnetic resonance\n(MRI) dataset. Through an extensive evaluation, we show that self-ensembling\ncan indeed improve the generalization of the models even when using a small\namount of unlabelled data.\n