A Robust Pose Optimization Scheme With Spatial Geometry Awareness for Hybrid Bronchoscopic Navigation

Interventional bronchoscopy is a minimally invasive procedure used to diagnose and treat peripheral pulmonary nodules. Bronchoscope navigation is one of the crucial technologies for guiding the biopsy pose with precision. However, electromagnetic (EM) navigation is vulnerable to respiratory motion and tissue displacement, while visual navigation of continuous frames can become unstable in blurred scenes. To overcome these challenges, we propose a robust two-stage hybrid navigation framework that integrates spatial geometry awareness with a hybrid update block. Our spatial-aware network leverages the strength of normal and gradient losses to deal with small structural fluctuations and stepped edges. It is trained on a synthetic scenario dataset and can be generalized to in vivo scenes. Using the initial pose from EM keypoint registration, a hybrid update block employs the spatial-aware network as a reliable and content-rich feature comparator to perform efficient optimal pose search with swarm intelligence. Positional and orientational accuracy of our method has been verified and compared with recent state-of-the-art (SOTA) methods in phantom and animal experiments. Both quantitative and qualitative results demonstrate that we provide a promising method for improving the accuracy and robustness of bronchoscopic navigation.