Accurate and efficient segmentation of liver tumors is essential for precise diagnosis, treatment planning, and monitoring of patients. To address these limitations, we proposed a novel framework called Deep Transfer Attention Network (DTAN) that integrates transfer learning and attention mechanisms for automated liver tumor segmentation. As a feature extractor, the DTAN model leverages a pre-trained convolutional neural network (CNN) to learn high-level representations from liver MRI images. To capture fine-grained spatial dependencies and emphasize tumor regions of interest, we introduce an attention mechanism that adaptively weights local features based on their relevance to the liver tumor segmentation task. We evaluate the model using metrics such as Hausdorff distance, Dice coefficient, specificity and sensitivity. The combination of transfer learning and attention mechanisms enables the extraction of discriminative features and enhanced understanding of spatial context, leading to more accurate and reliable liver tumor segmentation results. The proposed framework holds significant potential in supporting radiologists and clinicians in making timely and informed decisions for liver tumor diagnosis, treatment planning, and patient management.