Image segmentation is important for quantitative analysis in many medical applications. Although deep learning methods generate segmentation with high accuracy, the model has to be trained for different organs and image modalities. These training processes are very time-consuming and labor-intensive because of the requirements of drawing masks manually. In this project, we present a dual neural network method trained using one set of 3D MRI data from a single subject. We demonstrate the feasibility of using a single 3D dataset to train a dual neural network for kidney segmentation, which greatly reduce the burden of drawing masks for large datasets.