<p>Liver cancer is the tenth most prevalent cancer, as reported by the 2024 global cancer statistics. Despite advancements in imaging technology, the accurate identification and segmentation of hepatic structures, particularly blood vessels and tumors in computed tomography (CT) scans continue to pose significant challenges due to the low contrast and complex anatomical features of liver tissues. Traditional manual segmentation methods are impractical and time-consuming for clinical workflows. For addressing these limitations, this study proposes an engineering-driven solution named MultiResMorphNet, a novel deep learning architecture designed for detecting liver tumors. The approach begins with a robust preprocessing pipeline incorporating resizing, normalization and adaptive histogram equalization (AHE) to enhance local contrast and make tumor boundaries more distinguishable. The core of the solution lies in the MultiResMorphNet model, which combines multi-resolution analysis with morphological operations within a U-Net framework, enabling the network to effectively capture features. This architecture enhances segmentation accuracy by integrating low-resolution context with high-resolution detail, resulting in more precise localization of tumors. Implemented using Python, the proposed model achieves a high accuracy of 96.70%, outperforming several other approaches demonstrating its ability for real-time, automated liver tumor segmentation in clinical practice.</p>