SpReg: An Autonomous Image-To-Patient Registration Framework for Robotic Bronchoscopy
Dingpei Han, Tianqi Huang, Fang Chen,∗
AI summary
Problem
Clinical robotic bronchoscopy registration currently relies on manual, expert-driven navigation that is labor-intensive, time-consuming, and highly variable, creating a bottleneck for widespread clinical adoption.
Approach
The framework automates registration by fusing real-time bronchoscopic images with physician eye-tracking data to guide autonomous path planning, then aligns the recorded 3D trajectory with preoperative CT airway centerlines using a direction-aware iterative closest point algorithm.
Key results
- First fully automated CT-to-patient registration framework for robotic bronchoscopy without manual intervention
- Simulation trajectory error maintained below 1.6 mm during autonomous motion
- In vivo porcine tests show 19% reduction in navigation error and smoother trajectories compared to manual operation
- Final registration accuracy matches manual methods with no statistically significant difference
Why it matters
It substantially reduces surgeon workload and enhances procedural safety and efficiency, accelerating the clinical adoption of autonomous robotic bronchoscopy for pulmonary disease diagnosis and treatment.
Abstract
Robotic bronchoscopy offers transformative po- tential for the precise diagnosis and treatment of pulmonary diseases, yet its clinical adoption is bottlenecked by the challenge of rapidly and accurately registering preoperative CT images to the patient’s anatomy. Current methods, which rely on manual expert operation, are laborious and time-intensive with a steep learning curve, posing inherent risks to patients due to the lack of navigational support. Here, we present SpReg, an autonomous spatial registration framework that, for the first time, enables a robotic bronchoscope to perform autonomous driving and registration without manual intervention. The SpReg framework leverages a deep learning network that uniquely incorporates physicians’ eye-tracking data as prior information. This allows the system to identify key anatomical regions, using this as a foundation for autonomous path planning. The system then drives the robot to autonomously navigate along multi-level bronchial centerlines, recording its three-dimensional path, which is subsequently aligned with the preoperative CT model to complete the registration. Simulation experiments demonstrate that the trajectory error of SpReg during motion is below 1.6 mm. In vivo experiments in a porcine model further show that, compared to manual operation, SpReg produces smoother motion trajectories and reduces navigation error by 19% (2.1 mm vs. 2.6 mm). Notably, its final registration accuracy shows no statistically significant difference from the manual method. These findings demonstrate that SpReg has the potential to substantially reduce the surgeon’s workload while enhancing procedural safety and efficiency, paving the way for the development of more advanced human-robot collaborative intelligent surgical systems.