Precise segmentation of liver tumors is vital for computer-aided diagnosis, treatment planning, and monitoring disease progression. In this study, we propose an Attention Feature Fusion U-Net (AFF-U Net) for automatic segmentation of liver tumors from volumetric CT scans. The dataset, consisting of NIfTI liver volumes, was normalized and resized to a uniform resolution. The AFF-U-Net employs attention gates in the decoder to selectively emphasize tumor regions while reducing false positives in low-contrast areas. The network was trained using a combination of binary cross-entropy loss and Dice coefficient optimization. Experimental results demonstrate high segmentation performance, achieving a Dice coefficient of 0.9574, IoU of 0.9402, precision of 0.8828, recall of 0.8753, and accuracy of 0.9986. Computed tomography (CT) and magnetic resonance imaging (MRI) provide detailed three dimensional liver images, enabling tumor localization and assessment of growth. However, manual segmentation is time-consuming and prone to errors, with inter-observer variability and low contrast or irregular tumors often causing inconsistencies. further confirm accurate tumor delineation, underscoring the model’s potential for clinical applications.

Liver tumor segmentation, AFF-U-Net, Attention Gate, Deep Learning, CT Scan, Medical Image Analysis.