Abstract: Magnetic resonance imaging (MRI) is widely used in brain imaging. To better identify brain tissue and structure, brain MRI image segmentation algorithms have been developed, but most of the existing advanced segmentation models have high computational complexity and long computational times, which make them difficult to use widely. To facilitate clinical use, lightweight algorithms are required to improve efficiency and real-time performance. First, in this paper, the research background and significance of lightweight algorithms, and the development history of lightweight segmentation algorithms, are summarized. Next, lightweight segmentation technology is introduced, and the residual module, deep separable convolution, attention mechanism and hierarchical design method are discussed. The lightweight effect of the above methods is compared and analyzed under the framework of U-net and Transformer. Finally, current challenges of lightweight algorithms are illustrated, and future development direction is proposed.