Automating Vascular Shunt Insertion with the dVRK Surgical Robot

Karthik Dharmarajan, Will Panitch, Muyan Jiang, Kishore Srinivas, Baiyu Shi, Yahav Avigal, Huang Huang, Thomas Low, Danyal Fer, Ken Goldberg

TL;DR

Researchers at UC Berkeley and UCSF have developed an autonomous pipeline for vascular shunt insertion—a critical trauma surgery task—using the da Vinci Research Kit (dVRK). By combining deep-learning-based perception with a specialized "chamfer tilt and screw motion" maneuver, the system achieves up to a 95% success rate in inserting shunts into deformable vessel phantoms.

What Problem Does This Paper Address?

In emergency trauma scenarios, such as on a battlefield or at a major accident site, restoring blood flow to a limb is a race against time. Surgeons often use a vascular shunt—a flexible tube—to bridge a damaged blood vessel and temporarily restore circulation.

Automating this is incredibly difficult for several reasons:

  1. Deformability: Blood vessels are soft and change shape when touched.
  2. Occlusion: The robot's own "hands" (grippers) often block the camera's view of the vessel rim.
  3. Precision: Cable-driven surgical robots like the da Vinci suffer from "hysteresis" (mechanical lag and slippage), making millimeter-level accuracy a challenge.
  4. Tight Tolerances: The shunt is often nearly as wide as the vessel itself, leaving almost no room for error during insertion.

Methods

The authors propose a multi-stage pipeline called AVSI (Automated Vascular Shunt Insertion):

  • Perception via "LUV": To train their vision system, they used a clever trick called Invisible UV Fluorescent Indicators. They painted the vessel rims with UV paint, allowing them to automatically generate perfect "ground truth" labels by toggling between visible and UV light. They trained an asymmetric U-Net to segment the vessel rim from standard RGB images.
  • Circle Fitting: Once the rim is segmented, the system uses RANSAC (Random Sample Consensus) to fit a 3D circle to the noisy point cloud data, determining the vessel's center, orientation, and radius.
  • Bimanual Coordination:
    • Arm 1 (The Dilator): Grasps the vessel rim and pulls it outward to stretch the opening.
  • Arm 2 (The Inserter): Performs a complex insertion move. Instead of pushing straight in, it uses a "chamfer tilt" (angling the shunt to catch the edge) followed by a "screw motion" (rotating the shunt while pushing). This rotation helps the shunt clear the vessel walls and "settle" inside.

Datasets

The researchers created and utilized:

  • A custom training set: 3,200 UV/visible light image pairs of vessel phantoms in various orientations.
  • A validation set: 805 image pairs with corresponding 3D depth data.
  • Public Contribution: They released a dataset of 1,500 RGB images and segmentation masks to the research community.

Key Insights and Ideas

  • The "Screw" is Secret Sauce: The most significant insight is that a simple "push" often fails because the vessel rim buckles. The "screw motion" (counter-clockwise rotation during translation) acts as a mechanical stabilizer, allowing the shunt to overcome friction and tight tolerances.
  • Active Dilation is Mandatory: For shunts that are close to the vessel's diameter, insertion is impossible without active dilation. The success rate for large shunts dropped from 80% to 0% when the dilation step was removed.
  • Color Invariance: By training on grayscale versions of their data, they proved their perception model isn't just "looking for yellow" (the color of their phantom), making it more likely to generalize to real, blood-stained tissue in the future.

Key Results

The system was tested on vessel phantoms at various angles (0°, 15°, and 30°):

  • 8mm Shunt (High clearance): Achieved a 95% success rate with an average completion time of ~13.7 seconds.
  • 14mm Shunt (Tight clearance): Achieved an 80% success rate at 0° inclination.
  • Robustness: The system maintained high success rates even when the vessel was tilted at 15°, though performance dropped for the larger shunt at 30° due to the limitations of the initial "tilt" angle.
  • Speed: The entire autonomous procedure takes less than 15 seconds, which is highly efficient for emergency medical applications.

Limitations and Future Work

While promising, the authors note several hurdles before this hits the operating room:

  • Scale and Realism: The phantoms used were 50-100% larger than real human vessels, and latex balloons don't perfectly mimic the slippery, multi-layered nature of real arteries.
  • Static Environment: The vessel was held by fixed mechanical fixtures. In a real surgery, a human assistant might be holding the vessel, requiring the robot to use "visual servoing" to track a moving target.
  • Complexity: Real shunt procedures require "ligature knots" to secure the tube in place; currently, the robot only handles the insertion.

The team suggests that future work will focus on using more realistic tissue-like materials and integrating the system onto more mobile robotic platforms designed for field use.

Summary updated

Abstract

Vascular shunt insertion is a fundamental surgical procedure used to temporarily restore blood flow to tissues. It is often performed in the field after major trauma. We formulate a problem of automated vascular shunt insertion and propose a pipeline to perform Automated Vascular Shunt Insertion (AVSI) using a da Vinci Research Kit. The pipeline uses a learned visual model to estimate the locus of the vessel rim, plans a grasp on the rim, and moves to grasp at that point. The first robot gripper then pulls the rim to stretch open the vessel with a dilation motion. The second robot gripper then proceeds to insert a shunt into the vessel phantom (a model of the blood vessel) with a chamfer tilt followed by a screw motion. Results suggest that AVSI achieves a high success rate even with tight tolerances and varying vessel orientations up to 30°. Supplementary material, dataset, videos, and visualizations can be found at https://sites.google.com/berkeley.edu/autolab-avsi.
BibTeX
@inproceedings{dharmarajan2023automating,
  author = {Karthik Dharmarajan and Will Panitch and Muyan Jiang and Kishore Srinivas and Baiyu Shi and Yahav Avigal and Huang Huang and Thomas Low and Danyal Fer and Ken Goldberg},
  title = {{Automating Vascular Shunt Insertion with the dVRK Surgical Robot}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2023},
  url = {https://openreview.net/forum?id=FJz0ySxQ9Y},
}

Loading related content...

Figure