Medical image segmentation has become a significant focus for researchers, with UNet emerging as a particularly effective tool for precise analysis and detection of complex medical images. UNet’s capabilities have been widely documented, especially as enhancements to its architecture have led to significant improvements in performance. Various strategies exist to optimize UNet architecture, such as incorporating attention mechanisms, multi-scale features, residual connections, dense connectivity, and transformer-based enhancements. However, the simultaneous implementation of all these strategies can result in increased computational complexity and costs. This paper reviews studies on the application of different attention mechanisms across various stages of UNet to enhance segmentation performance. Models such as ResDSda-U-Net, DoubleU-NetPlus, and 3D-CU-Net demonstrate the effectiveness of attention mechanisms while keeping computational complexity low. These advancements are anticipated to enhance medical imaging technologies further, leading to better patient care and outcomes.