PaperlayerView exact revision on arXiv

Source-linked AI summary

Autonomous Robotic System with Optical Coherence Tomography Guidance for Vascular Anastomosis

Jesse Haworth, Rishi Biswas, Justin Opfermann, Michael Kam, Yaning Wang, Desire Pantalone, Francis X. Creighton, Robin Yang, Jin U. Kang, Axel Krieger

arXiv:2410.07493v1cs.ROcs.SYeess.SY

TL;DR

Small-vessel vascular anastomosis requires extensive training, and prior autonomous suturing systems had not been demonstrated in vascular surgery. µSTAR combines OCT sensing, camera-based missed-suture detection, and autonomous vessel manipulation to suture real tissue. In an ex vivo study, it placed 90% of sutures without human intervention and achieved outcomes described as competitive with experienced surgeons.

  • Problem

    Vascular anastomosis on vessels smaller than one millimeter demands extensive training, while prior autonomous suturing systems had not been demonstrated in vascular surgery.

    Source: §I · p. 1

  • Method

    µSTAR combines OCT-guided tissue sensing, camera-based missed-suture detection, and an autonomous suturing workflow with vessel manipulation.

    Source: §I · p. 2

  • Results

    90% of sutures were placed without human intervention, and µSTAR's outcomes were competitive with experienced surgeons; its suture-spacing COV was 30% versus Surgeon 2's 62%.

    Source: §IV · p. 8, D. Ex Vivo Comparison Study · p. 8

  • Takeaways & Limitations

    The ex vivo results mark an advance toward fully autonomous vascular anastomosis and may reduce reliance on highly skilled surgeons for complex procedures.

    Source: §V · p. 9

    A graduate student tied the knots manually after suturing, introducing variability in the final anastomosis, particularly in bubble-leak measurements.

    Source: §IV · p. 8

Abstract

from arXiv · show

Vascular anastomosis, the surgical connection of blood vessels, is essential in procedures such as organ transplants and reconstructive surgeries. The precision required limits accessibility due to the extensive training needed, with manual suturing leading to variable outcomes and revision rates up to 7.9%. Existing robotic systems, while promising, are either fully teleoperated or lack the capabilities necessary for autonomous vascular anastomosis. We present the Micro Smart Tissue Autonomous Robot (micro-STAR), an autonomous robotic system designed to perform vascular anastomosis on small-diameter vessels. The micro-STAR system integrates a novel suturing tool equipped with Optical Coherence Tomography (OCT) fiber-optic sensor and a microcamera, enabling real-time tissue detection and classification. Our system autonomously places sutures and manipulates tissue with minimal human intervention. In an ex vivo study, micro-STAR achieved outcomes competitive with experienced surgeons in terms of leak pressure, lumen reduction, and suture placement variation, completing 90% of sutures without human intervention. This represents the first instance of a robotic system autonomously performing vascular anastomosis on real tissue, offering significant potential for improving surgical precision and expanding access to high-quality care.

I. Introduction

Vascular anastomosis on small vessels demands extensive training, while existing autonomous suturing systems had not been demonstrated in vascular surgery. µSTAR is presented as an autonomous system for ex vivo vascular anastomosis that integrates sensing, error correction, and vessel manipulation.

I. Introduction

  • Vessels smaller than one millimeter require extensive training, limiting the number of surgeons able to perform these procedures and access to them.More than six million reconstructive surgeries were performed in the United States in 2020, many involving vascular anastomosis.Source: §I · p. 1
  • Manual suturing depends heavily on surgeon skill; studies report revision rates up to 7.9%, with complications often linked to inconsistent suture placement.Source: §I · p. 1
  • Prior autonomous suturing systems automated tasks such as knot tying or suture placement but had not been demonstrated in vascular surgery or addressed microvessel manipulation and deformation.Source: §I · p. 2
  • The stated requirements include accurate tissue sensing and needle driving, real-time error correction, vessel manipulation, limited human intervention, and demonstrated feasibility.µSTAR combines an OCT-and-camera suturing tool, tissue classification, missed-suture detection, and an autonomous workflow.Source: §I · p. 2
  • µSTAR is presented as the first robotic system to meet the stated capabilities for autonomous vascular anastomosis on ex vivo tissue.The study compares its performance with experienced surgeons.Source: §I · p. 2

II. Materials and Methods

µSTAR combines an OCT- and camera-equipped suturing tool with MAPS vessel handling, enabling an autonomous workflow that places sutures around the vessel.

A. Robotic Suturing Tool with Integrated Optical Coherence Tomography (OCT) Fiber-Optic Sensor and Microcamera

  • The modified Endo360 tool drives a curved needle along a fixed circular path and rigidly integrates OCT and a microcamera relative to the needle.The OCT provides real-time tool-to-tissue feedback; the camera images the suture site before and after placement to detect missed sutures.Source: A. Robotic Suturing Tool with… · pp. 2–3
  • The integrated OCT and microcamera provide sensing and visualization for autonomous vascular anastomosis.Source: A. Robotic Suturing Tool with… · p. 3

B. Vessel Manipulation and Suturing Workflow

  • MAPS was upgraded for autonomous control and real-tissue handling, with coated vessel holders, smoother mechanisms, and positioning-code changes to reduce slippage and overshoot.Source: B. Vessel Manipulation and Suturing… · p. 3
  • MAPS rotates aligned vessels during suturing, following the clinical pattern of suturing one side and flipping the clamp to access the other.The holders grip the vessels, and the carriages rotate in tandem to maintain alignment while minimizing excessive torque.Source: B. Vessel Manipulation and Suturing… · p. 3
  • The tool passes the needle outside-in through the right vessel half and inside-out through the left; after the suture is pulled through and cut, MAPS rotates the vessel for the next stitch.The illustrated sequence shows the resulting stitch before rotation.Source: B. Vessel Manipulation and Suturing… · p. 3, Figure 2 · p. 3

C. Tissue Detection and Classification with OCT

µSTAR uses OCT signal profiles and template matching to classify material and locate the vessel edge for suture placement; failed or false edge detection can require manual adjustment.

C. Tissue Detection and Classification with OCT

  • OCT intensity profiles distinguish tissue, air, and nitinol; the system saves a signal segment one millimeter past the detected tissue threshold as a vessel-wall template.Source: C. Tissue Detection and Classification… · p. 4
  • The algorithm smooths and normalizes the template and incoming signal, then compares them using RMSE to classify tissue or nitinol.Normalization sets each signal's maximum intensity to 1, helping robustness to attenuation from distance, moisture, or light deflection.Source: C. Tissue Detection and Classification… · p. 4
  • When the scan identifies air or nitinol, it halts and marks the OCT fiber position as the vessel edge for setting needle bite depth.Source: C. Tissue Detection and Classification… · p. 4
  • After a failed edge scan, the system retries up to three times; persistent failure or a false edge requires manual robot adjustment before suturing.Algorithm thresholds were tuned using more than 24 ex vivo scans.Source: C. Tissue Detection and Classification… · p. 4

D. Missed Suture Detection

The system detects missed sutures from paired microcamera images so it can respond during the procedure. A modified ResNet-50 provides the binary classification.

D. Missed Suture Detection

  • Missed-suture detection identifies failures to engage tissue and provides a mechanism for the system to respond during the procedure.The system uses a ResNet-50 model for detection.Source: D. Missed Suture Detection · pp. 4–5
  • 540 before-and-after image pairs were collected, with nearly three times as many successful as missed sutures; training batches balanced the classes.Images were captured using the system’s microcamera.Source: D. Missed Suture Detection · p. 4
  • The classifier concatenates before-and-after RGB images into a six-channel input and maps its features through 230 logits to two binary-classification outputs.The added block also includes batch normalization and 50% dropout.Source: D. Missed Suture Detection · p. 4
  • Training used cross-entropy loss, Adam at a 0.0003 learning rate, data augmentation, and early stopping after 30 epochs without loss reduction.Augmentations included rotations, flips, color jitter, pixel dropout, and affine transformations.Source: D. Missed Suture Detection · p. 4

E. Software Architecture and Autonomous Control Workflow

The software architecture coordinates hardware and control components across ROS and ROS2. The workflow combines vessel-edge scanning, suturing, missed-suture checks, and vessel rotation, with user involvement at specified steps.

E. Software Architecture and Autonomous Control Workflow

  • The architecture uses ROS for the KUKA arm and Endo360 tool, and ROS2 for newer components including OCT, MAPS, the microcamera, and the controller.The ROS interfaces were already available for the arm and tool; ROS2 was selected for newer components partly for its real-time performance and longer-term support.Source: E. Software Architecture and… · p. 5, Figure 4 · p. 5
  • The ROS2 µSTAR Controller coordinates component tasks through services, while a ROS/ROS2 bridge links the KUKA arm and Endo360 with ROS2 components.The OCT Node scans tissue, the MAPS Node rotates vessels, and the Endo360 Node manages suturing.Source: E. Software Architecture and… · p. 5
  • The user loads the vessels, pulls through and cuts each suture, and ties off at the end; the robot uses calibrated offsets to align the needle with the vessel edge.The z offset is measured with the tool flush against paper; x and y offsets are recorded after puncturing the paper.Source: E. Software Architecture and… · p. 5
  • For each stitch, the system captures a before-image, scans the right vessel edge with OCT, places a suture, captures an after-image, and prompts the user if a missed stitch is detected.The sequence is repeated for the left vessel, then MAPS rotates the vessel to the next suturing location.Source: E. Software Architecture and… · p. 5
  • The authors describe the combined ROS and ROS2 architecture as enabling autonomous task management while reducing the need for human intervention.The architecture is also described as maintaining flexibility for future enhancements.Source: E. Software Architecture and… · p. 5

A. Vessel Positioning System Evaluation

The MAPS evaluation measured vessel-holder repeatability and assessed the forces associated with gripping and puncturing an ex vivo vessel.

A. Vessel Positioning System Evaluation

2) Grip Force and Puncture Testing:

1) Repeatability Testing:

  • The left and right stages averaged 44.9 degrees (SD 2.8, n=16) and 45.3 degrees (SD 2.2, n=16) per rotation, respectively; full-rotation error averaged 1.9±1.3 degrees (n=4).The full-rotation error is reported for both holders together.Source: 1) Repeatability Testing · p. 6

B. Tissue Classification Testing and Evaluation

In ex vivo testing, µSTAR classified vessel edges and transition materials, trained a missed-suture detector, and completed anastomoses with most sutures placed autonomously. In comparisons with surgeons, it showed no significant bubble-leak difference and lower suture-spacing variation than Surgeon 2, while taking longer per stitch.

B. Tissue Classification Testing and Evaluation

  • Across 49 scans of six artery samples, the system identified tissue edges in 89.8% and correctly labeled transition material in 88.6% of samples.The OCT scans classified the transition material as nitinol or air.Source: B. Tissue Classification Testing and… · p. 6

C. Missed Suture Detection Evaluation

D. Ex Vivo Comparison Study

  • µSTAR placed 90% of sutures autonomously, with no significant bubble-leak differences from surgeons and lower spacing COV% than Surgeon 2 (30% vs 62%).The system’s average bite depth was 1.54±0.22 mm, with an average error of 0.39 mm from the 1.5 mm target.Source: D. Ex Vivo Comparison Study · pp. 7–8
  • Vessel slippage along the holder and later rotation under suture tension led to one crossed stitch, with the final suture overlapping the first.Source: D. Ex Vivo Comparison Study · p. 8
  • Time per stitch averaged 352.7 seconds for µSTAR versus 141.2 seconds for surgeons; significant within-group differences also occurred for lumen reduction and time per stitch.The comparison involved three surgeons and µSTAR, each performing anastomoses on five porcine femoral arteries; a graduate student tied the robot’s sutures.Source: D. Ex Vivo Comparison Study · pp. 7–8

E. Statistical Analysis

Statistical comparisons used one-way ANOVA, followed by Tukey HSD for group p-values below 0.05; COV% differences were calculated separately.

E. Statistical Analysis

  • One-way ANOVA compared individual surgeons with µSTAR, Tukey HSD followed group p-values below 0.05, and MedCalc was used for COV% differences.Statistically significant results were labeled in Figs. 8A, 8B, and 8C.Source: E. Statistical Analysis · p. 8

IV. Discussion

In ex vivo testing, µSTAR produced outcomes competitive with experienced surgeons and placed 90% of sutures without human intervention. The discussion identifies constraints involving suture size, knot-tying, tissue handling, speed, and vessel diameter.

IV. Discussion

  • µSTAR achieved ex vivo outcomes competitive with experienced surgeons; Fig. 9 compares lumen reduction, bubble leak, and average time per stitch.Source: §IV · p. 8, Figure 9 · p. 8
  • 90% of sutures were placed without human intervention, with OCT sensing and ResNet50 missed-suture detection supporting placement and error correction.Source: §IV · p. 8
  • µSTAR remained competitive in lumen reduction despite using 3-0 braided polyester, larger than typical for this anastomosis.The suturing tool does not currently support smaller suture sizes.Source: §IV · p. 8
  • A graduate student tied the sutures after robotic placement, and knot-tying inconsistencies could affect bubble-leak results.Source: §IV · p. 8
  • Vessel slippage along the holder and holder rotation led to one crossed stitch.The slippage was likely due to the needle’s curved path through tissue.Source: §IV · pp. 8–9
  • µSTAR was slower than surgeons, primarily because of robot movement during vessel-edge scanning; manual knot-tying also added time.Source: §IV · pp. 8–9
  • The system is currently limited to anastomosis on vessels approximately five millimeters in diameter because of tool and holder size.Source: §IV · p. 9

V. Conclusion

µSTAR performed vascular anastomosis autonomously with outcomes competitive with experienced surgeons, placing 90% of sutures without human intervention. The conclusion identifies smaller suture compatibility and autonomous knot-tying as areas for improvement.

V. Conclusion

  • µSTAR performed vascular anastomosis with outcomes competitive with experienced surgeons and placed 90% of sutures without human intervention.Source: §V · p. 9
  • The study marks an advancement toward fully autonomous vascular anastomosis and could reduce reliance on highly skilled surgeons for complex procedures.Source: §V · p. 9
  • Future efforts will address smaller suture compatibility, fully autonomous knot-tying, and miniaturization for smaller vessels.Source: §V · p. 9

Related papers

Loading 2410.07493v1…