Source-linked AI summary
Predictive Multi-Landmark OCT Tracking for Increased Motion Robustness
Konrad Reuter, Suresh Guttikonda, Chaitali Uday Karekar, Christian Betz, Alexander Schlaefer
TL;DR
OCT-based markerless tracking is limited by velocity, especially when multiple landmarks are acquired sequentially for 6D pose estimation. The paper introduces predictive multi-landmark tracking using global pose propagation, achieving RMSEs below 1 mm up to 100 mm/s and nine landmarks. Its applicability to real-world tracking remains to be demonstrated.
Problem
Sequential multi-landmark OCT tracking has limited trackable velocity because more landmarks reduce each landmark's effective acquisition rate.
Method
The method uses individual landmark trackers and Kabsch-based global pose estimates to predict all landmark positions at the current and next timestep.
Results
RMSEs remain below 1 mm for velocities up to 100 mm/s across up to nine consecutively tracked landmarks.
Takeaways & Limitations
Global pose propagation improves OCT tracking robustness under rapid motion across the evaluated velocities and landmark counts.
Takeaways & Limitations
The results use simulated, well-controlled motion, and applicability to a physical real-world OCT tracking system remains un demonstrated.
Abstract
from arXiv · showhide
Optical coherence tomography is a promising modality for markerless motion tracking due to its high spatial resolution and inherent depth perception. However, existing OCT-based tracking approaches are limited in terms of trackable velocity, particularly when multiple landmarks are tracked sequentially for 6D pose estimation. In this work, we present a predictive tracking approach that propagates positional updates between multiple tracked landmarks to obtain a global pose prediction. This enables more robust tracking under high velocities. Our results demonstrate RMSEs below 1 mm for velocities up to 100 mm/s and up to nine consecutively tracked landmarks, highlighting the potential of global motion propagation and prediction for improving the robustness of OCT-based tracking.
1 Introduction
OCT offers markerless tracking with high spatial resolution and depth information, but sequential multi-landmark tracking remains velocity-limited. The proposed approach predicts global pose changes by propagating motion information across landmarks.
- 1 Introduction: OCT supports markerless tracking through high spatial resolution, reduced susceptibility to reflections, and inherent depth information.Its limited field of view requires active adjustment of lateral scanning and depth reference positions.
- 1 Introduction: Existing OCT template matching requires sufficient overlap between consecutive volumes, so large relative displacements impair reliable motion estimation.Prior OCT tracking demonstrated submillimeter accuracy, but trackable velocities remained limited.
- 1 Introduction: Sequential acquisition reduces each landmark’s effective acquisition rate, lowering the maximum trackable velocity as more landmarks are tracked.Nearby landmarks often exhibit similar motion, allowing motion information from one landmark to predict neighboring landmarks.
- 1 Introduction: The method propagates positional updates across multiple landmarks to estimate a global pose change and predict their next positions.The approach is evaluated on OCT scans of porcine skin under simulated motion profiles.
2 Methods
The method uses one tracker per landmark while assuming a shared rigid structure, then estimates pose changes from sequential landmark updates. Evaluation combines calibrated OCT acquisition data with simulated motion and compares independent and predictive tracking across landmark counts.
- 2 Methods: Each landmark is tracked with an individual MOSSE instance, while the framework assumes all landmarks belong to the same rigid structure.The framework can incorporate trackers other than MOSSE.
- 2 Methods: The Kabsch transformation between successive landmark configurations is divided by the number of landmarks to approximate per-step translational and rotational pose changes.The transformation represents motion accumulated over N tracked-landmark acquisition steps.
- 2 Methods: The evaluation simulates motion by virtually moving subvolumes generated from linear combinations of real OCT volumes at an assumed 832 Hz volume rate.Tracker latency and modeled galvo and reference-arm motion are included.
- 2 Methods: Baseline independent tracking is compared with the proposed method for 3, 5, 7, and 9 landmarks over 30-second trajectories.Motion amplitudes are limited to 40 mm laterally and 20 mm in depth.
- 2 Methods: Tracking performance is evaluated with RMSE, using the centroid of tracked landmark positions for a single position estimate.The study also investigates the overlap required for reliable MOSSE tracking.
3 Results
Reliable lateral tracking extends to approximately 11 voxels before errors rise substantially. The predictive method maintains reliable tracking up to 100 mm/s across landmark counts, with errors exceeding 1 mm only for nine landmarks.
- 3 Results: Lateral displacements up to approximately 11 voxels can be tracked reliably, but errors increase substantially beyond that range.From 13 voxels onward, the median error becomes comparable to the induced displacement.
- 3 Results: Reliable tracking remains possible at velocities up to 100 mm/s independently of landmark count, while errors exceed 1 mm only for 9 landmarks.The baseline errors increase with both velocity and the number of simultaneously tracked landmarks.
4 Discussion
The proposed global-pose prediction substantially expands trackable velocities and supports all evaluated landmark counts, while performance varies with landmark count and remains bounded by evaluation conditions.
- Baseline behavior: The baseline’s maximum trackable velocity decreases as more landmarks are tracked because sequential acquisition lengthens each landmark’s update interval.The resulting displacement can exceed the allowable tracking threshold at lower target velocities.
- Overall performance: The proposed method substantially increases trackable velocities and enables successful tracking across all evaluated speeds and landmark counts.It propagates one landmark’s positional update through a shared global pose estimate to predict all landmarks.
- Landmark-count effects: Tracking with 5 and 7 landmarks consistently produces lower errors than tracking with 3 or 9 landmarks.Fewer landmarks make individual errors more influential, whereas nine landmarks increase the delay between observations and make motion information less reliable.
- Limitations: The results were generated with simulated motion and well-controlled dynamics, so evaluation on the physical OCT tracking system remains necessary.The implementation also models velocity in voxels per update, although explicit temporal modeling may improve robustness and prediction accuracy.
5 Conclusion
The paper presents predictive motion estimation for multi-landmark OCT tracking using global pose changes estimated with the Kabsch algorithm. The results are promising, but applicability to a real-world tracking system remains to be demonstrated.
- Conclusion: Kabsch-based global pose estimation improves multi-landmark OCT tracking robustness under rapid motion.The approach performs predictive motion estimation using tracked landmarks.
- Conclusion: The approach’s applicability in a real-world tracking system remains to be demonstrated.