FSA-Net: Fractal-driven Synergistic Anatomy-aware Network for Segmenting White Line of Toldt in Laparoscopic Images

Accurate automatic segmentation of the White Line of Toldt (WLT) is crucial for guiding colorectal cancer surgeries and improving patient outcomes. However, the complex anatomical structures and low signal-to-noise ratio involved in relevant regions of WLT pose significant challenges to existing segmentation models. Recent studies highlight fractal dimension as a powerful tool for analyzing the complexity of topological structures, offering an effective approach to representing anatomical features in medical images. Building on its success, we present the first well-annotated laparoscopic WLT segmentation (LTS) dataset and propose FSA-Net, a fractal-driven synergistic anatomy-aware network, specially designed for laparoscopic WLT segmentation. Specifically, FSA-Net consists of two core modules: the local texture-aware convolution (LTC) module and the fractal-guided anatomy-consistent attention (FAA) module. The LTC module adaptively adjusts the convolutional kernel offsets based on fractal dimensions to capture intra-anatomical features, while the FAA module employs a fractal-driven key-value pair filtering strategy to enhance the modeling of correlations across inter-anatomical structures. Extensive experimental results validate the effectiveness of our method.