Stop Holding Your Breath: CT-Informed Gaussian Splatting for Dynamic Bronchoscopy Reconstruction

Forum topic · 小凯 · 2026-05-02

Summary

This arXiv paper (2604.28179) proposes a CT-informed Gaussian splatting framework for dynamic bronchoscopic navigation that eliminates the need for breath-hold protocols. Respiratory motion deforms the airway by 5-20 mm, causing CT-to-body divergence that limits endoscopic localization. The method exploits paired inspiration-expiration CT scans already acquired for planning to define a patient-specific deformation space of the airway. By registering these scans, respiratory motion is reduced to a single scalar breathing phase per frame, constraining reconstruction to anatomically observed configurations. This representation is embedded in a mesh-anchored Gaussian splatting framework, where a lightweight estimator infers breathing phase directly from endoscopic RGB, enabling continuous deformation-aware reconstruction across the full respiratory cycle without breath-holds or external sensors. The authors also introduce RESPIRE, a physics-based bronchoscopy simulation pipeline providing per-frame ground truth for geometry, pose, breathing phase, and deformation. Experiments show geometrically faithful reconstructions, over 20x faster training, and 1.22 mm target localization accuracy, within the 3 mm clinically relevant tolerance.

Paper Overview

Field: CV/Medical Authors: Anna E. S Published: 2026-04-30 arXiv: 2604.28179

Abstract (English translation)

Bronchoscopic navigation relies on registering endoscopic video to a preoperative CT scan, but respiratory motion deforms the airway by 5-20 mm, creating CT-to-body divergence that limits localization accuracy. In practice, this is mitigated through breath-hold protocols, which attempt to match the intraoperative anatomy to a static CT, but are difficult to reproduce and disrupt clinical workflow.

The authors propose eliminating the need for breath-hold protocols by leveraging patient-specific respiratory modeling. Paired inspiration-expiration CT scans, already acquired for planning purposes, implicitly define a patient-specific deformation space of the breathing airway. By registering these scans, respiratory motion is reduced to a single scalar breathing phase per frame, constraining all reconstructions to lie within anatomically observed configurations.

This representation is embedded into a mesh-anchored Gaussian splatting framework, where a lightweight estimator infers the breathing phase directly from endoscopic RGB, enabling continuous, deformation-aware reconstruction throughout the respiratory cycle — without breath-holds or external sensing.

Key Contributions

Breath-hold-free navigation: Patient-specific respiratory modeling removes the need for hard-to-reproduce breath-hold protocols.
Scalar breathing phase: Airway deformation is parameterized as a single per-frame scalar derived from registering paired inspiration-expiration CT scans.
Mesh-anchored Gaussian splatting with a lightweight RGB-based phase estimator for deformation-aware reconstruction.
RESPIRE benchmark: A physics-based bronchoscopy simulation pipeline providing per-frame ground truth for geometry, pose, breathing phase, and deformation.

Results

On RESPIRE, the method achieves geometrically faithful reconstructions, over 20x faster training, and 1.22 mm target localization accuracy — within the 3 mm clinically relevant tolerance.

---

*Auto-collected on 2026-05-02*