Accurate segmentation of MR brain tissue is a crucial step for diagnosis,surgical planning, and treatment of brain abnormalities. However,it is a time-consuming task to be performed by medical experts. So, automatic and reliable segmentation methods are required. How to choose appropriate training dataset from limited labeled dataset rather than the whole also has great significance in saving training time. In addition, medical data labeled is too rare and expensive to obtain extensively, so choosing appropriate unlabeled dataset instead of all the datasets to annotate, which can attain at least same performance, is also very meaningful. To solve the problem above, we design an automatic segmentation method based on U-shaped deep convolutional network and obtain excellent result with average DSC metric of 0.8610, 0.9131, 0.9003 for Cerebrospinal Fluid (CSF), Gray Matter (GM) and White Matter (WM) respectively on the well-known IBSR18 dataset. We use bootstrapping algorithm for selecting the most effective training data and get more state-of-the-art segmentation performance by using only 50% of training data. Moreover, we propose a strategy of MR brain tissue images' suggestive annotation for unlabeled medical data based on the modified U-net. The proposed method performs fast and can be used in clinical.