HANS-Net is a novel segmentation framework for accurate liver and tumor segmentation in abdominal CT images, which are challenging due to complex anatomy, variation in tumor appearance, and limited annotation data. It integrates hyperbolic convolution for hierarchical geometric representation, a wavelet-inspired decomposition module for multi-scale texture learning, a biologically motivated synaptic plasticity mechanism for adaptive feature enhancement, and implicit neural representation branches to model fine and continuous anatomical boundaries. It also incorporates uncertainty-aware Monte Carlo dropout to quantify prediction confidence and lightweight temporal attention to improve slice-to-slice consistency without sacrificing efficiency. Extensive evaluation on the LiTS dataset shows that HANS-Net achieves an average Dice score of 93.26%, IoU of 88.09%, an average symmetry surface distance (ASSD) of 0.72 mm, and a volume overlap error (VOE) of 11.91%. Cross-dataset validation on 3D-IRCADb-01 dataset achieves an average Dice of 87.45%, IoU of 80.30%, ASSD of 1.525mm, and VOE of 19.71%, demonstrating strong generalization performance on different datasets. These results confirm the effectiveness and robustness of HANS-Net in providing anatomically consistent, accurate, and reliable liver and tumor segmentation.
