The mid-level features combining the local features into a global image representation are representative and discriminative,which can serve as an image representation.Many successful models for medical image segmentation propose efficient methods to learn mid-level features,such as sparse coding technology combined with spatial pyramid matching (SPM),dictionary learning,neural network,etc.The application of mid-level feature improves the performance of segmentation algorithm.This paper introduces the medical image segmentation methods based on mid-level features and prospects the feature research work.