ESD-VesNet: uncertainty-aware vessel segmentation network for endoscopic submucosal dissection with hard negative mining - Summary - MDSpire

ESD-VesNet: uncertainty-aware vessel segmentation network for endoscopic submucosal dissection with hard negative mining

  • By

  • Mengya Xu

  • Ming Chen

  • Zhen Li

  • Chaoyang Lyu

  • An Wang

  • Rulin Zhou

  • Chuanhao Zhao

  • Jiaxun Xiang

  • Tsz Chun Wong

  • Hossein Farahnaki

  • Sobhan Zamani Kiasari

  • Tong Wu

  • Zimeng Su

  • Yile Zeng

  • Ruijing Wen

  • Xiaohan Shang

  • Yi Mu

  • Kezhen Lin

  • Yidong Zhang

  • Hongliang Ren

  • July 3, 2026

  • 0 min

Share

Objective:

To develop an AI-assisted framework for accurate vessel detection and segmentation in endoscopic submucosal dissection (ESD) procedures, addressing challenges related to intraoperative bleeding.

Approach:
  • Dataset Creation: Introduced the ESD-Vessel dataset with 44 high-definition surgical procedures, comprising 2401 annotated vessel frames and 708 hard negative frames.
  • Framework Development: Proposed ESD-VesNet, a vessel segmentation framework based on the Segment Anything Model 3 (SAM3) with uncertainty-aware learning and hard negative mining.
Key Findings:
  • The ESD-Vessel dataset is the first to incorporate negative samples to reduce false positives during training.
  • ESD-VesNet achieves superior segmentation quality with high vessel detection rates and low false-positive rates.
Interpretation:

The proposed methods aim to improve the detection and segmentation of vascular structures.

Limitations:
  • The study relies on a specific dataset from a single institution, which may limit generalizability.
  • Challenges remain in the dynamic surgical environment that may affect segmentation accuracy.
Conclusion:

The integration of uncertainty-aware learning and hard negative mining in vessel segmentation could significantly improve procedural safety in ESD.

Original Source(s)

ESD-VesNet: uncertainty-aware vessel segmentation network for endoscopic submucosal dissection with hard negative mining

  • by Mengya Xu, Ming Chen, Zhen Li, Chaoyang Lyu, An Wang, Rulin Zhou, Chuanhao Zhao, Jiaxun Xiang, Tsz Chun Wong, Hossein Farahnaki, Sobhan Zamani Kiasari, Tong Wu, Zimeng Su, Yile Zeng, Ruijing Wen, Xiaohan Shang, Yi Mu, Kezhen Lin, Yidong Zhang, Hongliang Ren

  • July 3, 2026

Related Content

Robotic-assisted endoscopic submucosal dissection: a scoping review of preclinical and early clinical evidence

Lightweight deep learning model for gastrointestinal precancerous lesion screening with attention enhancement

Laparoscopic and open feeding jejunostomy in upper gastrointestinal pathology: a single-centre cohort study