In this model, to solve the problem of segmentation results lack of clear boundary, 
we propose a boundary weighted loss function, which can make the network focus more on boundary information 
and accurately detect and segment boundary. Besides, it also can keep the continuity and smooth of the boundary. 
To solve the problem of small dataset in deep learning based medical image segmentation,
 we propose a deeply supervised domain adaptation model for transferring network trained
 in source domain towards target domain to leverage available data from other domains. 
The proposed model learns a transformation from other datasets to the medical image datasets domain space in an adversarial learning manner.
 Furthermore, to efficiently extract the 3D spatial contextual information of volumetric data for feature representations, 
we propose a novel 3D deep feature representation network structure.