PaperlayerView exact revision on arXiv

Source-linked AI summary

Contrast-Free Autonomous Navigation of Untethered Endovascular Microrobots Using Single-Plane Fluoroscopy

Husnu Halid Alabay, Tuan-Anh Le, Ping Wang, Hakan Ceylan

arXiv:2608.30220v1cs.RO

TL;DR

Single-plane fluoroscopy complicates autonomous microrobot navigation because it provides only a 2D projection, while repeated contrast imaging increases procedural burden. VISTA uses vascular topology and a patient-specific digital twin to guide navigation, reducing navigation time, corrective commands, and X-ray exposure relative to manual control.

  • Problem

    Single-plane fluoroscopy lacks depth information for autonomous navigation, motivating contrast-sparing approaches because contrast use increases cumulative procedural burden.

  • Method

    VISTA uses a vascular digital twin and projected navigation milestones to convert fluoroscopic robot observations into topology-constrained navigation states and actuation commands.

  • Results

    Up to 57% lower X-ray exposure was achieved relative to manual control, alongside reductions in navigation time and corrective actuation commands.

  • Takeaways & Limitations

    Known vascular topology and a patient-specific digital twin can support contrast-free autonomous navigation of untethered endovascular microrobots.

  • Takeaways & Limitations

    The vascular model is static and does not capture intraoperative anatomical changes; external magnetic-actuator accessibility is also a physical constraint.

Abstract

from arXiv · show

Reliable three-dimensional (3D) navigation of magnetically actuated untethered microrobots remains a major barrier to clinical translation. X-ray fluoroscopy is the standard real-time imaging modality for endovascular procedures, but single-plane fluoroscopy provides only a two-dimensional (2D) projection, eliminating depth information and complicating autonomous navigation. Recovering this information through biplane imaging or repeated contrast-enhanced angiography increases procedural complexity, radiation exposure, or contrast burden. Here, we introduce VISTA (Virtual Integration for Spatial Tracking and Autonomy), a digital twin framework enabling contrast-free autonomous navigation under single-plane fluoroscopy. VISTA reconstructs vascular anatomy as a 3D digital twin, discretizes the vessel centerline into navigation milestones, and assigns the detected 2D robot position to the nearest projected milestone. Consecutive milestones define the local vessel orientation used to generate magnetic actuation commands, converting single-plane fluoroscopic observations into topology-constrained navigation states without requiring contrast injection during navigation. VISTA is demonstrated across anatomically distinct vascular phantoms under continuous flow and within the inferior vena cava of a live rat in vivo. Compared with conventional fluoroscopic human-in-the-loop control, VISTA reduced navigation time by up to 62%, corrective actuation commands by up to 98%, and radiation exposure by up to 57%. These results establish VISTA as a digital twin-guided framework for contrast-free autonomous navigation of untethered endovascular microrobots using widely available single-plane fluoroscopy.

1 Department of Physiology and Biomedical Engineering, Mayo Clinic, Scottsdale, Arizona, USA

VISTA addresses the depth ambiguity of single-plane fluoroscopy by using known vascular topology and a 3D digital twin for contrast-free autonomous navigation. The framework was evaluated in vascular phantoms and a live rat model, with improved navigation efficiency and reduced operator intervention relative to manual control.

  • Clinical motivation: Single-plane fluoroscopy collapses 3D vascular anatomy into a 2D projection, making a robot’s position along the vessel trajectory ambiguous.This ambiguity complicates closed-loop magnetic navigation, especially in complex geometries with overlapping projections.
  • VISTA framework: VISTA represents the known 3D vascular centerline as discrete milestones and maps each 2D robot detection to a corresponding state along that trajectory.Consecutive milestones provide the local vessel orientation used to configure rotating magnetic actuation.
  • Contrast-free navigation: VISTA uses fluoroscopy to track the radiopaque robot rather than repeatedly visualize the vessel lumen, eliminating contrast administration during navigation.The digital twin retains vascular context throughout the procedure.
  • Validation: VISTA was validated in anatomically distinct vascular phantoms under continuous flow and in the inferior vena cava of a live rat.The experiments demonstrated autonomous closed-loop navigation with reduced operator intervention.
  • Implication: The digital-twin approach is presented as a scalable framework for clinically deployable untethered endovascular microrobots using single-plane fluoroscopy.Its scope is tied to known vascular topology and a registered patient-specific anatomical model.

RESULTS

VISTA constructs a patient-specific vascular digital twin from segmented 3D angiographic data and extracts an ordered vessel centerline for navigation. The centerline is discretized into 3D milestones that define discrete navigation states.

  • Digital-twin construction: A patient-specific vascular digital twin is reconstructed from a segmented 3D angiographic model and corresponding projection images.The vascular mesh is skeletonized to extract the vessel centerline.
  • Centerline representation: The vessel centerline is represented as a spatial curve parameterized by arc length.This curve provides the geometric basis for milestone placement.
  • Navigation states: The centerline is discretized into 3D milestones that define discrete navigation states along the vascular topology.Each milestone corresponds to a sampled point on the centerline.

was detected at the image-plane coordinate

VISTA assigns the detected robot to the nearest projected vessel milestone and uses neighboring milestones to estimate local vessel direction. The resulting piecewise-constant axis balances navigation accuracy with actuation and imaging constraints.

  • Milestone localization: The detected robot position is mapped to the nearest projected milestone in the registered digital twin.The active milestone pair then supplies the local vessel direction for magnetic control.
  • Actuation control: Consecutive milestones define the local vessel tangent and the rotation axis of the commanded magnetic field.The rotating field lies orthogonal to the local vessel orientation and generates torque for helical propulsion.
  • Event-driven control: VISTA updates the commanded rotation axis only when the nearest milestone changes, reducing unnecessary robot-arm motion.The axis remains piecewise constant between milestone transitions.
  • Control constraint: Frequent orientation updates triggered robot-controller safety limits and prevented reliable locomotion during preliminary testing.Small changes in desired magnetic-axis orientation could require comparatively large Cartesian robot-arm motions.
  • Resolution limits: Navigation-state resolution is bounded by approximately 0.1 mm manipulator displacement and approximately 0.3 mm fluoroscopic spatial resolution.Increasing resolution beyond these limits adds command burden without providing equivalent control information.
  • Actuation tolerance: 18.26° and 25.79° were the angular tolerances for 95% and 90% magnetic-actuation retention, respectively.These tolerances link allowable actuation loss to milestone spacing.
  • Spacing selection: 7.77 mm and 12.93 mm were the maximum milestone spacings for at least 95% and 90% magnetic-actuation retention, respectively.The spacing criterion connects vascular geometry, magnetic actuation, and navigation-state discretization.

Accurate Detection, Localization and Real-Time Position Estimation

VISTA registers reconstructed vascular anatomy with fluoroscopic images and uses robust EndoBot detection to assign each observation to a discrete 3D navigation state. The detector maintained high accuracy across tested orientations, and control updates fit within the fluoroscopic imaging interval.

  • Digital-twin registration: Each vascular phantom is reconstructed volumetrically and incorporated into Unity as a vessel-specific digital twin.Fiducial markers align the physical phantom and virtual model in position and scale.
  • 3D state estimation: The detected 2D EndoBot position is matched to the nearest projected centerline milestone to obtain a discrete 3D navigation state.The associated local orientation is derived from the active milestone pair and defines the magnetic-field rotation axis.
  • Detection robustness: 99.2% overall detection rate was achieved, with greater than 90% detection accuracy at every tested orientation.The single EndoBot detector reliably identified the robot from 0° to 90° in 15° increments.

Real-Time Robotic Execution and Path-Following Validation

VISTA’s validated perception-to-command pipeline translated fluoroscopic detections and milestone transitions into accurate, timely robotic actuation. It navigated distinct vascular geometries autonomously under flow and in a rat IVC, with propulsion adapting to flow direction and in vivo conditions.

  • Robotic validation: The manipulator accurately executed 10-30 cm translations and 0°-90° rotations with positional and angular errors below 0.5 mm.
  • Real-time execution: 20 ms end-to-end latency completed each localization and control cycle before the next fluoroscopic frame.
  • Path-following validation: All three vascular phantoms supported fully autonomous inlet-to-outlet navigation without modifying the underlying navigation algorithm or control parameters.
  • Flow-dependent performance: Against flow, navigation required 93.3% more time and 66.7% more robot-arm movements while reducing mean traversal velocity by 48.3%.The centerline advancement per movement decreased from approximately 18.0 to 10.8 mm.
  • In vivo validation: In the rat IVC, autonomous navigation covered an approximately 162-mm centerline path in both flow directions without human intervention.The path took 15 seconds with nine unique robot-arm movements in-flow and 29 seconds with 15 movements against flow.
  • In vivo control adaptation: Rotational actuation alone was insufficient for consistent translation under some in vivo conditions, so a margin-based translational adjustment increased magnetic gradient force.The adjustment addressed effects potentially associated with local blood flow and animal-specific vascular geometry.

Autonomous Navigation Outperforms Human Fluoroscopic Control

VISTA outperformed novice and expert manual fluoroscopic control in a standardized 3D helical phantom, reducing navigation time, movements, and X-ray exposure. Additional tests showed that performance depended on vascular geometry and magnetic-field alignment.

  • Standardized benchmark: VISTA completed navigation in 80.7 s using 2.67 movement actions and 1.02 mGy, versus 213.7 s, 120.3 actions, and 2.35 mGy for novices.The expert required 152.3 s, 101.7 movement actions, and 1.93 mGy.
  • Statistical comparison: Overall differences among the three control conditions were significant for total navigation time, X-ray exposure, and movement count.
  • Directional comparison: VISTA significantly improved all three outcomes during forward propulsion relative to both manual operators.During backward propulsion, it reduced time and exposure relative to the novice and movement count relative to both manual conditions.
  • Robustness testing: Under improper curvature alignment, novice performance deteriorated and the expert could not complete navigation because of EndoBot deformation, whereas VISTA maintained successful navigation.
  • Geometry and actuation effects: In a less tortuous phantom, VISTA reduced navigation time from 80.7 to 66 s and X-ray exposure from 1.02 to 0.63 mGy.The effects were geometry-dependent, and increasing rotational speed provided no clear benefit during manual control.

DISCUSSION

The discussion presents VISTA as a topology-constrained alternative to unconstrained 3D pose reconstruction: a patient-specific digital twin supplies spatial context and local vessel orientation while single-plane fluoroscopy tracks the robot. The framework reduces manual intervention and radiation exposure, but remains bounded by static anatomy, actuator reachability, and evaluation mainly along individual vessel segments.

  • Digital-twin representation: Known vascular topology converts limited 2D fluoroscopic observations into anatomically permissible navigation states and supports magnetic control.
  • Digital-twin representation: Consecutive centerline milestones define local vessel orientation and automatically update the rotating magnetic-field axis as the robot progresses.
  • Actuation-aware discretization: Milestone spacing trades geometric fidelity against robotic control efficiency: frequent updates increase manipulator motion, whereas larger intervals increase angular mismatch.
  • Practical implications: Compared with representative novice and experienced manual operators, VISTA reduced navigation time, corrective actuation commands, and X-ray exposure.The comparison is a controlled benchmark rather than a general comparison with clinical operator performance.
  • Practical implications: Effective propulsion depends strongly on maintaining the magnetic actuation axis’s relationship with local vascular geometry, particularly in tortuous trajectories.The digital-twin geometry provides a systematic means of maintaining this relationship, which is increasingly difficult to reproduce manually.
  • Limitations and future directions: The current digital twin is static and does not capture intraoperative changes in vessel geometry or physiology, including deformation from cardiac and respiratory motion.
  • Limitations and future directions: VISTA does not guarantee that every desired actuation configuration is physically achievable because human-scale workspace is constrained by depth, tortuosity, kinematics, clearance, and imaging-system interference.
  • Imaging and navigation: Separating real-time robot tracking from vascular visualization enables closed-loop, contrast-free navigation with single-plane fluoroscopy without biplane imaging or repeated contrast administration.

3D angiography and phantom vessel preparation

The target vessel was reconstructed from rotational angiography into a segmented 3D vascular mesh that formed VISTA’s digital-twin anatomy. Silicone phantom vessels were fabricated in varied geometries for three-dimensional navigation evaluation.

  • 3D angiography: 3D rotational angiography acquired 512 projections over approximately 200° and reconstructed them into a volumetric dataset.
  • Digital-twin anatomy: The target vessel was segmented into a 3D vascular mesh that served as the anatomical basis for VISTA localization and navigation.
  • Spatial registration: VISTA associated each detected 2D EndoBot position with the projected vascular centerline to estimate its 3D navigation state and local vessel orientation.
  • Phantom preparation: Silicone phantom vessels were designed and fabricated using 3D-printed sacrificial molds, silicone elastomer, and acetone dissolution of ABS filaments.
  • Phantom preparation: Phantom Models 2 and 3 were configured into different geometries to evaluate navigation under varying three-dimensional vessel structures.

Creation and evaluation of object detection model

The system detected the EndoBot from X-ray and RGB images using YOLOv8, with online augmentation expanding training variability. Detection outputs were integrated with Unity and robotic control for real-time actuation.

  • Object detection: YOLOv8 was trained on 817 X-ray and RGB images divided into 562 training, 131 validation, and 124 test images.
  • Object detection: Online augmentation applied randomized transformations during training, exposing the model to 28,100 augmented image presentations across 50 epochs.
  • Robotic control: The KUKA arm received real-time external motion commands and updated the external magnet pose using the FRI connection.
  • System integration: The detection pipeline published environmental and EndoBot tracking information to Unity, which converted inputs into estimated position and orientation.
  • Robotic control: Position and orientation targets were interpolated smoothly, while predefined thresholds constrained positional and rotational errors.

Unity setup and experimental fixes

Unity represented the virtual magnet with configurable milestone offsets so its trajectory could preserve the desired magnetic relationship to the EndoBot and vessel. Performance comparisons used three independent trials and predefined statistical tests.

  • Unity setup: A separate magnet-specific milestone set was offset from the vessel milestones to configure magnet position and orientation relative to the EndoBot.
  • Unity setup: Unity parameters could be modified through a control panel to adapt the virtual setup to different phantom geometries and experimental configurations.
  • Experimental evaluation: Performance metrics summed forward and backward navigation measurements across three independent standard-condition trials.
  • Statistical analysis: Data were reported as mean ± standard deviation, with one-sided unpaired Student’s t-tests assuming equal variances and significance defined as p < 0.05.

navigation. (C) EndoBot–milestone association and milestone initialization. The detected

VISTA localized the robot by associating fluoroscopic detections with vessel milestones, using consecutive milestones to estimate orientation and update magnetic actuation. The framework was evaluated in multiple vascular phantoms, ex vivo flow, and a live rat, with milestone spacing tied to torque retention.

  • Milestone localization: The detected EndoBot position was mapped to the nearest centerline milestone, and the active index advanced when the robot crossed milestone midpoints.
  • Actuation control: Consecutive milestones defined local vessel direction and updated the desired magnetic-field orientation during navigation.
  • Milestone spacing: A 95% torque-retention threshold corresponded to a maximum milestone spacing of 7.77 mm, whereas 90% retention corresponded to 12.93 mm.
  • Localization validation: Validation showed agreement between single-plane fluoroscopic detections and digital-twin estimates of the EndoBot’s 3D position and orientation.
  • Phantom navigation: VISTA was evaluated across three umbilical-vein-derived vascular phantoms with geometry-specific centerlines and milestone spacing.
  • Ex vivo and in vivo evaluation: Fully autonomous closed-loop navigation was demonstrated in ex vivo flow and during forward and backward navigation within the inferior vena cava of a live rat.
  • Performance comparison: Manual and VISTA navigation were compared using total navigation time, robot-arm movement actions, and cumulative X-ray exposure across three trials per condition.
Loading 2608.30220v1…