Landmark Detection Uncertainty as a Reliability Weight for Robust Landmark-based 2D/3D Pelvic Pose Estimation

Yehyun Suh, Brayden Schott, Chou Mo, John Ryan Martin, Daniel Moyer

Keywords: Uncertainty-Weighted Pose Estimation, Landmark-based 2D/3D Registration, Monte Carlo Dropout, Epistemic Uncertainty Modeling

Abstract: Landmark-based 2D/3D pelvis registration is vulnerable to noisy or ambiguous landmark detections in fluoroscopy, which can destabilize downstream pose estimation. We present an uncertainty-aware registration framework that models epistemic uncertainty in predicted landmarks and incorporates it directly into the Perspective-n-Point formulation. Using Monte Carlo dropout within a U-Net detector, we compute sample-specific per-landmark reliability estimates using the variance of multiple stochastic forward passes. These reliability estimates guide two complementary strategies: continuous weighting, which integrates uncertainty into a weighted PnP optimization, and discrete selection, which removes the most uncertain landmarks during inference. We evaluate the framework on both CT-derived synthetic fluoroscopy and real fluoroscopy from DeepFluoro. Our experiments show that uncertainty provides a principled mechanism for identifying unreliable landmarks and stabilizing pose estimation, enabling more robust registration and establishing a foundation for uncertainty-guided image-guided surgical workflows.

Primary Subject Area: Image Registration

Secondary Subject Area: Uncertainty Estimation

Registration Requirement: Yes

Reproducibility: https://github.com/yehyunsuh/Uncertainty-Aware-Pelvic-Pose-Estimation

Visa & Travel: No

Read CFP & Author Instructions: Yes

Originality Policy: Yes

Single-blind & Not Under Review Elsewhere: Yes

LLM Policy: Yes

Midl Latex Submission Checklist: Ensure no LaTeX errors during compilation., Replace NNN with your OpenReview submission ID., Includes \documentclass{midl}, \jmlryear{2026}, \jmlrworkshop, \jmlrvolume, \editors, and correct \bibliography command., Did not override options of the hyperref package., Did not use the times package., Use the correct spelling and format, avoid Unicode characters, and use LaTeX equivalents instead., Any math in the title and abstract must be enclosed within $...$., Did not override the bibliography style defined in midl.cls and did not use \begin{thebibliography} directly to insert references., Avoid using \scalebox; use \resizebox when needed., Included all necessary figures and removed *unused* files in the zip archive., Removed special formatting, visual annotations, and highlights used during rebuttal., All special characters in the paper and .bib file use LaTeX commands (e.g., \'e for é)., No separate supplementary PDF uploads., Acknowledgements, references, and appendix must start after the main content.

Submission Number: 132