Xincheng Yao

Laparoscopic hepatectomy demands real-time identification of critical vascular structures, including Glisson's pedicle, hepatic veins, and transient bleeding points, to enable pre-emptive vessel clamping and prevent life-threatening haemorrhage. These targets are frequently obscured and deformable, and in ultra-long procedures spanning thousands of frames they further suffer extreme scale imbalance and intermittent visibility, making memory-based propagation prone to temporal drift where a single corrupted entry cascades forward irreversibly. We present VigilSAM3, built on SAM3's dual-system architecture, which introduces a hierarchical memory firewall that guards memory quality at pixel, class, and frame levels. A causal-Transformer scheduling agent selects among Trust, Correct, and Reset actions, with its policy refined via Group Relative Policy Optimization (GRPO) using class-aware rewards. We further introduce Hepa-Vein, the first multi-class large-scale hepatectomy surgical video semantic segmentation dataset. Experiments on Hepa-Vein and DSAD demonstrate state-of-the-art performance, with pronounced gains on small, transient vascular structures.

High-resolution segmentation of hepatic vasculature in laparoscopic hepatectomy videos is essential for intraoperative guidance. Built on a frame-by-frame annotated dataset of 35 long hepatectomy videos with 11,442 high-resolution frames, HRVVS embeds a pretrained visual autoregressive (VAR) model into different layers of the hierarchical encoder as prior information to counter downsampling degradation, and employs a dynamic memory decoder on a multi-view segmentation network to limit redundancy across frames. Experiments show that HRVVS outperforms state-of-the-art methods on hepatic vasculature segmentation.

Laparoscopic liver resection (LLR) is challenging due to the complex and variable intrahepatic vascular anatomy, limited surgical field of view, and lack of tactile feedback. These difficulties can be mitigated through accurate real-time recognition of the anatomical structures shown in the laparoscopic view. In this work, we develop a video segmentation AI model for real-time anatomy recognition in LLR, and validate its performance on laparoscopic videos of hepatectomy. The proposed model recognizes target anatomical structures, including the liver, gallbladder, hepatoduodenal ligament, and hepatic pedicle, in real time during surgery, and demonstrates potential for assisting surgical decision-making and improving patient safety.