6DOF Needle Pose Estimation for Robot-Assisted Vitreoretinal Surgery

Mingchuan Zhou, Hao Xing, Abouzar Eslami, Kai Huang, Caixia Cai, Chris P. Lohmann, Nassir Navab, Alois C. Knoll, M. Ali Nasseri

Introduction

In the realm of microsurgery, vitreoretinal (VR) procedures represent one of the most daunting challenges. Surgeons must manipulate tissues as thin as 200µm, often requiring a precision of 20µm for delicate tasks such as retinal vein cannulation. Given that natural human hand tremors typically hover around 100µm, these operations are physically and mentally exhausting. While robotic assistance can eliminate tremors, it introduces a new problem: the need for the robot to "see" and track surgical tools with sub-micron accuracy in real-time.

The Precision Challenge in Vitreoretinal Surgery

Traditional surgical microscopes provide 2D views that struggle with depth perception and poor internal illumination. While Optical Coherence Tomography (OCT) offers 3D volumetric data, processing this information to track a moving needle is computationally intensive. Standard algorithms often fall into "local optima"—mathematical traps where the system incorrectly identifies the needle's position. To overcome these hurdles, researchers have developed a high-precision 3D navigation system that achieves sub-10µm accuracy by combining OCT with a specialized "Shift-Rotate" Iterative Closest Point (SR-ICP) algorithm.

A Novel 3D Navigation Pipeline

The proposed system utilizes a two-stage pipeline to estimate the needle's 6-degree-of-freedom (6DOF) pose. The researchers employed the iRAM!S ophthalmic robot and a Carl Zeiss RESCAN 700 intraoperative OCT to test their approach on ex-vivo pig eyes.

Experimental setup (a) and schematic of the surgical environment (b) showing the Remote Center of Motion (RCM) control.

Robust Segmentation and SR-ICP

The first stage of the pipeline is Robust Segmentation. The system identifies the needle within the 3D OCT volume using a voting mechanism based on geometric features. Because a needle's cross-section in an OCT B-scan resembles a half-ellipse, the algorithm fits ellipses to the data and clusters points belonging to the needle, effectively filtering out noise and reflections from the retina.

The second stage is the Shift-Rotate ICP (SR-ICP). Standard ICP algorithms search across all six degrees of freedom simultaneously, which is slow and error-prone. By leveraging the known cylindrical geometry of the surgical needle, the researchers reduced the 6DOF problem to a 2DOF optimization: shifting along the longitudinal axis and rotating around it. This dimensionality reduction significantly improves both speed and reliability.

The pose estimation process: (a) input OCT image, (b) segmented point cloud identifying the retina (1), needle (2), and shadow (3), and (c) the final estimated pose.

Experimental Results and Performance

The system was evaluated using 150 data samples from ex-vivo pig eyes, comparing the algorithm's output against a high-fidelity CAD model of the needle. Key findings include:

  • Micron-Level Precision: The system achieved a position accuracy of within 10µm with 95% confidence, surpassing the 20µm requirement for retinal procedures.
  • High Speed: The Levenberg-Marquardt (LM) version of the SR-ICP algorithm processed data in approximately 335.4 ms, which is 5 times faster than standard ICP methods.
  • Rotational Accuracy: Mean errors for the needle roll angle were between 0.08° and 0.36°, well within the tolerances for beveled needle insertion.

Limitations and Future Outlook

The primary bottleneck remains the OCT capture time, which takes approximately 500ms per volume. This latency could be problematic in live surgeries where the eye or needle might move during the scan. Future research will focus on improving capture speeds, developing automated path planning for the robot, and moving toward in-vivo trials to test the system against complex tissue interactions and blood flow.

Summary updated

Abstract

Vitreoretinal (VR) surgery is typical microsurgery with delicate and complex surgical procedures. The vision-based navigation for robot-assisted VR surgery has not been fully exploited because of the challenges that arise from illumination, high precision, and safety assessments. This paper presents a novel method to estimate the 6DOF needle pose specifically for the application of robotic intraocular needle navigation using optical coherence tomography (OCT) volumes. The key ingredients of the proposed method are (1) 3D needle point cloud segmentation in OCT volume and (2) needle point cloud 6DOF pose estimation using a modified iterative closest point (ICP) algorithm. To address the former, a voting mechanism with geometric features of the needle is utilized to robustly segment the needle in OCT volume. Afterward, the CAD model of the needle point cloud is matched with the segmented needle point cloud to estimate the 6DOF needle pose with a proposed shift-rotate ICP (SR-ICP). This method is evaluated by the existing ophthalmic robot on ex-vivo pig eyes. The quantitative and qualitative results are evaluated and presented for the proposed method.
BibTeX
@article{zhou2019dof,
  author = {Mingchuan Zhou and Hao Xing and Abouzar Eslami and Kai Huang and Caixia Cai and Chris P. Lohmann and Nassir Navab and Alois C. Knoll and M. Ali Nasseri},
  title = {{6DOF Needle Pose Estimation for Robot-Assisted Vitreoretinal Surgery}},
  journal = {IEEE Access},
  year = {2019},
  doi = {10.1109/access.2019.2912327},
}
Figure