VeNet: a lightweight neural network for efficient brain vessel segmentation in endovascular robotic surgery

Area of Science:

• Medical Robotics
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Telesurgery leverages AI and robotics for remote surgical procedures.
• Accurate patient-specific anatomical models are crucial for AI-assisted surgery.
• Current vessel segmentation methods struggle with limited data and computational demands.

Purpose of the Study:

• To develop a lightweight neural network (VeNet) for efficient 3D segmentation of tubular structures like blood vessels.
• To address the challenges of data scarcity and computational cost in vascular segmentation for robotic endovascular surgery.
• To create a resource-efficient foundation for vascular segmentation and simulation workflows.

Main Methods:

• Developed VeNet, a lightweight neural network with a matrix-based operator and ~6,000 trainable parameters.
• Utilized efficient training and inference on standard CPU hardware.
• Generated a large, semi-automatically annotated brain vessel magnetic resonance angiography dataset from the IXI cohort with expert review.

Main Results:

• VeNet demonstrated robust segmentation performance in low-data regimes, suitable for scenarios with limited manual annotation.
• The model enables efficient training and near-real-time inference on standard hardware.
• A comprehensive dataset for vascular segmentation was created, facilitating research and development.

Conclusions:

• VeNet provides a resource-efficient solution for 3D vascular segmentation, particularly in data-limited settings.
• The developed dataset and model support patient-specific vascular reconstruction and interactive simulation in robotic surgery.
• VeNet's integration into a robotic platform enhances capabilities for endovascular interventions.