Learning-Based 3D Trajectory Generation for Automated Nasal Endoscopy

The anatomical complexity of the nasal cavity has led most nasal endoscopy examinations to remain reliant on manual or semi-automated control, which not only demands substantial physician time and effort but also limits clinical efficiency. To alleviate physician workload and enhance examination efficiency, this paper proposes an automated nasal endoscopy examination method based on trajectory generation network. Given a frontal facial image, 3DDFA-V3 is employed to reconstruct the corresponding facial point cloud. This point cloud is then processed by the proposed trajectory generation network to produce a 3D trajectory for nasal endoscopy examination. Finally, a robotic nasal endoscopy system executes based on the trajectory, completing the examination autonomously. Experimental results demonstrate that the proposed method completes the nasal examination in approximately 28 seconds, representing a significant improvement in clinical efficiency.