Online Topological Localization for Navigation Assistance in Bronchoscopy

TL;DR

This work tackles the challenge of navigating bronchoscopy by introducing an online, image-based topological localization approach that uses a generic bronchial tree model and requires no pre-operative CT scans. It combines a frame-based location classifier, trained solely on phantom data with a 5-level grayscale input, with a Bayesian localization framework and a branching-point detector to robustly estimate the bronchoscope's position online. The approach demonstrates strong real-data improvements over existing methods, achieving notable gains in top-1 and top-3 accuracy on real sequences while maintaining competitive phantom performance, thereby enabling navigation guidance and potential autonomous bronchoscopy. By avoiding CT-based registration and leveraging phantom-trained models, the method offers practical generalization, reduced setup, and lower data labeling costs, with broad applicability to other endoscopic procedures.

Abstract

Video bronchoscopy is a fundamental procedure in respiratory medicine, where medical experts navigate through the bronchial tree of a patient to diagnose or operate the patient. Surgeons need to determine the position of the scope as they go through the airway until they reach the area of interest. This task is very challenging for practitioners due to the complex bronchial tree structure and varying doctor experience and training. Navigation assistance to locate the bronchoscope during the procedure can improve its outcome. Currently used techniques for navigational guidance commonly rely on previous CT scans of the patient to obtain a 3D model of the airway, followed by tracking of the scope with additional sensors or image registration. These methods obtain accurate locations but imply additional setup, scans and training. Accurate metric localization is not always required, and a topological localization with regard to a generic airway model can often suffice to assist the surgeon with navigation. We present an image-based bronchoscopy topological localization pipeline to provide navigation assistance during the procedure, with no need of patient CT scan. Our approach is trained only on phantom data, eliminating the high cost of real data labeling, and presents good generalization capabilities. The results obtained surpass existing methods, particularly on real data test sequences.

Figures (4)

• Figure 1: Our pipeline determines the topological location of the camera, given the video feed observed up to the current frame (in green), with regard to a generic tree model of the airway (with nodes named as TRA, RMB, LMB,..., as done clinically). Examples of frames for some of the classes are shown in the diagram, highlighting the low inter-class variation.
• Figure 2: Overview of our proposed bronchoscopy topological localization approach. The current frame in the sequence goes through the frame-based CNN classifier and the branching-point detector. The output is then combined with the distributions of the previous frames to obtain the posterior distribution and the current frame localization in the bronchial tree.
• Figure 3: Examples of data used: images from a phantom visentini2017deep and real data. (a) Original RGB data as captured during the procedure. (b) 5-level grayscale image, obtained running k-means (k=5) on the RGB image, used as input for our frame-based classifier.
• Figure 4: Branching-point detector applied on three real examples: (a) input real image (b) result of intensity threhsolding to select the darkest pixel set $D$ (in red). (c) result of applying the connected components algorithm to group $D$-pixels into instances (each color segment represents a different instance $L_k$). The instances are filtered to keep only the largest ones, that correspond to lumen regions indicating different airway channels.