We developed active learning and semi-supervised learning techniques for reducing the labeling cost when training a deep neural network based segmentation or classification algorithm.