Proposed a novel architecture, Selective Attention UNet, for liver tumor segmentation in medical imaging applications. Our experimental results demonstrate that the proposed architecture outperforms several baseline models including FCN, UNet++, and SegNet in terms of accuracy and robustness metrics on the publicly available LiTS dataset. The Selective Attention UNet architecture leverages both low-level and high-level features by incorporating skip connections between the encoder and decoder pathways, and uses a selective attention mechanism to selectively focus on important features while suppressing irrelevant ones. Our results show that this architecture is able to accurately segment liver tumors while preserving important spatial information.