HANS-Net: Hyperbolic Convolution and Adaptive Temporal Attention for Accurate and Generalizable Liver and Tumor Segmentation in CT Imaging

TL;DR

This work tackles the challenge of accurate liver and tumor CT segmentation under limited labeled data by introducing HANS-Net, a framework that fuses hyperbolic convolution, wavelet-inspired multiscale decomposition, a synaptic plasticity mechanism, an implicit neural representation branch, adaptive temporal attention, and uncertainty estimation. The model achieves state-of-the-art performance on LiTS (Dice Liver 96.67%, Tumor 89.84%) and demonstrates strong cross-dataset generalization on AMOS 2022 (Dice 85.09%), while providing reliable pixel-level uncertainty maps and efficient inter-slice consistency. Ablation studies confirm additive gains from each module, with HC and INR delivering notable improvements and the full system reaching a mean Dice of 93.86% in LiTS evaluations. The approach offers anatomically coherent, high-precision segmentation with quantified confidence, signaling potential for clinical adoption and motivating future work on broader modality generalization and runtime optimization.

Abstract

Accurate liver and tumor segmentation on abdominal CT images is critical for reliable diagnosis and treatment planning, but remains challenging due to complex anatomical structures, variability in tumor appearance, and limited annotated data. To address these issues, we introduce Hyperbolic-convolutions Adaptive-temporal-attention with Neural-representation and Synaptic-plasticity Network (HANS-Net), a novel segmentation framework that synergistically combines hyperbolic convolutions 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 an implicit neural representation branch to model fine-grained and continuous anatomical boundaries. Additionally, we incorporate uncertainty-aware Monte Carlo dropout to quantify prediction confidence and lightweight temporal attention to improve inter-slice consistency without sacrificing efficiency. Extensive evaluations of the LiTS dataset demonstrate that HANS-Net achieves a mean Dice score of 93.26%, an IoU of 88.09%, an average symmetric surface distance (ASSD) of 0.72 mm, and a volume overlap error (VOE) of 11.91%. Furthermore, cross-dataset validation on the AMOS 2022 dataset obtains an average Dice of 85.09%, IoU of 76.66%, ASSD of 19.49 mm, and VOE of 23.34%, indicating strong generalization across different datasets. These results confirm the effectiveness and robustness of HANS-Net in providing anatomically consistent, accurate, and confident liver and tumor segmentation.

Figures (10)

• Figure 1: Overview of the proposed HANS-Net architecture that combines hyperbolic convolutions, wavelet decomposition, temporal attention, synaptic plasticity, implicit neural representation, and uncertainty estimator.
• Figure 2: The Hyperbolic Convolution Module first applies convolution, then flattens and transforms the input before passing it through a hyperbolic mapping block to obtain hyperbolic features.
• Figure 3: Work flow Temporal Attention Mechanism that processes input features through query-key-value extraction, attention weight calculation, and temporal weight modulation to generate output features
• Figure 4: The process of implicit neural representation, comprised of encoding coordinates with positional encoding, combining them with a latent feature vector, and passing the result through an MLP function to generate the output.
• Figure 5: The process of uncertainty estimator using Monte Carlo Dropout, where the input feature map undergoes spatial dropout, followed by inference, mean prediction, and uncertainty estimator