Core Concepts
The author presents the DD-VNB framework, integrating depth estimation and dual-loop localization for real-time bronchoscopic navigation, emphasizing speed and accuracy in clinical applications.
Abstract
The DD-VNB framework proposes a novel approach to bronchoscopic navigation by combining depth estimation and dual-loop localization. It addresses the challenges of generalization across patient cases without re-training, achieving superior speed and accuracy. The methodology involves knowledge-embedded depth estimation, ego-motion inference, and depth map registration for precise localization. Experimental results demonstrate significant improvements over existing methods in both phantom and patient data.
Stats
Localization achieves an accuracy of Absolute Tracking Error (ATE) of 4.7 ± 3.17 mm in phantom data and 6.49 ± 3.88 mm in patient data.
The frame-rate approaches video capture speed.
Monocular depth estimation outperforms SOTA.
Quotes
"The DD-VNB framework integrates two key modules: depth estimation and dual-loop localization."
"Our contributions include a generalizable VNB framework that eliminates the need for re-training."
"The proposed framework outperforms SOTA in localization accuracy."