MRI scans appearance significantly depends on scanning protocols and,\nconsequently, the data-collection institution. These variations between\nclinical sites result in dramatic drops of CNN segmentation quality on unseen\ndomains. Many of the recently proposed MRI domain adaptation methods operate\nwith the last CNN layers to suppress domain shift. At the same time, the core\nmanifestation of MRI variability is a considerable diversity of image\nintensities. We hypothesize that these differences can be eliminated by\nmodifying the first layers rather than the last ones. To validate this simple\nidea, we conducted a set of experiments with brain MRI scans from six domains.\nOur results demonstrate that 1) domain-shift may deteriorate the quality even\nfor a simple brain extraction segmentation task (surface Dice Score drops from\n0.85-0.89 even to 0.09); 2) fine-tuning of the first layers significantly\noutperforms fine-tuning of the last layers in almost all supervised domain\nadaptation setups. Moreover, fine-tuning of the first layers is a better\nstrategy than fine-tuning of the whole network, if the amount of annotated data\nfrom the new domain is strictly limited.\n