Towards a Videobronchoscopy Localization System from Airway Centre

Tracking

Carles S

anchez

, Antonio Esteban-Lansaque

, Agn

es Borr

, Marta Diez-Ferrer

, Antoni Rosell

and Debora Gil

Computer Vision Center, Computer Science Department, UAB, Bellaterra, Spain

Pneumology Unit, Hospital Universitari Bellvitge, IDIBELL, CIBERES, Barcelona, Spain

Keywords:

Video-bronchoscopy, Lung Cancer Diagnosis, Airway Lumen Detection, Region Tracking, Guided Bron-

choscopy Navigation.

Abstract:

Bronchoscopists use ﬂuoroscopy to guide ﬂexible bronchoscopy to the lesion to be biopsied without any kind

of incision. Being ﬂuoroscopy an imaging technique based on X-rays, the risk of developmental problems and

cancer is increased in those subjects exposed to its application, so minimizing radiation is crucial. Alternative

guiding systems such as electromagnetic navigation require speciﬁc equipment, increase the cost of the clinical

procedure and still require ﬂuoroscopy. In this paper we propose an image based guiding system based on

the extraction of airway centres from intra-operative videos. Such anatomical landmarks are matched to the

airway centreline extracted from a pre-planned CT to indicate the best path to the nodule. We present a

feasibility study of our navigation system using simulated bronchoscopic videos and a multi-expert validation

of landmarks extraction in 3 intra-operative ultrathin explorations.

1 INTRODUCTION

Lung cancer is one of the most incidental neoplasms

in Europe with around 437, 773 new cases estimated

for 2015

. Screening programs increase lung can-

cer early diagnosis and, thus, may signiﬁcantly reduce

the risk of death (Aberle et al., 2011). Screening pro-

grams are based on the detection of small pulmonary

lesions with low dose chest computed tomography

(CT) and its pathological conﬁrmation. Pathologi-

cal conﬁrmation requires either transthoracic needle

aspiration or an endoscopic examination. Transtho-

racic needle aspiration has some complications (Man-

hire et al., 2003), like pneumothorax (20%). Conven-

tional bronchoscopic diagnostic procedures are visu-

ally guided using radiating ﬂuoroscopy and render a

suboptimal 34% of positive results for lesions < 2 cm

(Donnelly and Edwin, 2012). New endoscopy tech-

niques (like virtual bronchoscopy or electromagnetic

techniques) are expensive, require either manual in-

tervention or special gadgets, only increase diagnos-

tic yield to 70% (Aberle et al., 2011), and still radiate

the patient. The 30% undiagnosed pulmonary lesions

need CT follow-up or futile surgery procedures such

globocan.iarc.fr

as thoracoscopies (Aberle et al., 2011). Diagnostic

yield could be improved reducing radiation and costs

if imaging technology could better detect and guide

the bronchoscopist to the target lesion.

A main limitation of ﬂexible bronchoscopy is the

difﬁculty to determine the correct pathway to periph-

erial lesions. CT Virtual Bronchoscopy (VB) can pre-

cede bronchoscopy to assess the optimal path using

virtual navigation. VB is a computer simulation of

the video bronchoscope image from the bronchoscope

camera (Reynisson et al., 2014) which is created from

the 3D CT volume, with the same view angle and

zoom settings. Despite different approaches to VB,

existing systems fail due to several aspects at plan-

ning and exploration time.

During exploration, VBs should accurately guide

the operator across the planned path to the biopsy

point. Standard protocols relying on ﬂuoroscopy

have a diagnostic yield around 40%, last 20 min

per intervention and require 5-10 min of repetitive

patient and medical staff radiation (Asano et al.,

2013). Existing alternatives like image based systems

(LungPoint, NAVI) or electromagnetic navigation (in-

ReachTM,SPinDriveA) are far from meeting clini-

cian expectations. Image systems are based on multi-

modal registration of CT virtual projections to bron-

352

SÃ ˛anchez C., Esteban Lansaque A., BorrÃ

as A., Diez-Ferrer M., Rosell A. and Gil D.

Towards a Videobronchoscopy Localization System from Airway Centre Tracking.

DOI: 10.5220/0006115803520359

In Proceedings of the 12th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2017), pages 352-359

ISBN: 978-989-758-225-7

choscopy frames and require manual intra-operative

adjustments of the guidance system (Eberhardt et al.,

2010). Electromagnetic systems require speciﬁc gad-

gets (Gildea et al., 2006), that alter the operating pro-

tocol and increase intervention cost.

The potential of image processing in enhancing

guiding capabilities has not been fully explored. To

display the correct position of the bronchoscope in the

CT-derived maps (structural maps of airways), scope

position should be tracked in videos. Some solutions

consist in hybrid algorithms (Lu

o et al., 2012) (fea-

ture tracking and multimodal registration) but most

of current solutions are based on multimodal registra-

tion (Reynisson et al., 2014) of CT virtual projections

to the actual videobronchosocpy frame (Mirota et al.,

2011). A main disadvantage is that the view from

the bronchoscope can be obscured by blood or mu-

cus, causing the tracking between video images and

CT to be disrupted. Another ﬂaw is the sensitivity

to common sudden changes in image appearance and

camera motion.

An alternative to image registration is the use

of anatomical landmarks representing airway geom-

etry to match CT planning to current navigation path.

Some authors (Hofstad et al., 2015) use airways cen-

treline extracted using an electromagnetic tracking

sensor on the tip of the bronchoscope. Another option

is to extract landmarks representing the intrinsic local

geometry of airways which allows its identiﬁcation

in, both, CT scans and videos without any extra de-

vice (S

anchez et al., 2015b). Such geometric features

can be encoded using a computational data structure

for path matching. Landmark extraction in interven-

tional videos is challenging due to the large variety of

image and motion artefacts and the unpredicted pres-

ence of endoscopic devices. Although recent works

anchez et al., 2014; S

anchez et al., 2015a) have de-

veloped reliable processing methods to extract endo-

luminal structures, they are designed for 2D image

analysis and might fail applied to videos.

We propose a codiﬁcation of airways as a binary

tree deﬁned by bronchial branches to match lumen

centres between CT and videobronchoscopy. Lu-

men centres are tracked in images using a multi-

scale approach based on maximally stable external

regions (MSER) (Matas et al., 2004). Tracked cen-

tres together with detection of branching points are

used to match the VB planing and the current ex-

ploration. The feasibility of our landmark based CT-

Video matching is tested on several video sequences

simulated on a VB interface platform. The capability

for intra-operative matching is checked by assessing

the quality of the tracked centres in ultrathin real ex-

plorations. These ﬁrst results show the potential of

our methodology to accurately reach most distal lev-

els in ultrathin navigation.

2 CT-VIDEO ANATOMICAL

MATCHING

The bronchial tree has a tubular geometry and, thus,

it is described as far as its central line (airway lu-

men centre) and walls (luminal area) are extracted.

In the case of bronchial navigation, the path can be

described using the lumen centre position and the

branching points of the bronchial tree structure.

Our CT-video path matching locates the current

position of the scope by comparing the bronchial tree

extracted from the CT used to plan the intervention to

a bronchial structure generated from the tracking of

lumen centres extracted from videobronchoscopic im-

ages during intervention time. Both anatomical struc-

tures can be computationally encoded by means of a

binary tree (S

anchez et al., 2015b) with nodes given

by the bronchial branching levels. The matching be-

tween CT-video bronchial structures is then given by

comparing the two binary trees.

2.1 Bronchial Anatomy Encoding from

the CT Scan

The whole bronchial tree to be matched to the current

exploration navigation path is encoded from the skele-

ton of a segmentation of the CT volume. The skeleton

is obtained using (Van Uitert and Bitter, 2007) with a

branch pruning based on its length. The 3D binary

volume containing the one-dimensional voxel skele-

ton is converted (Kerschnitzki and et al., 2013) into a

network graph described by nodes (skeleton branch-

ing points) and edges (branch joining two nodes) or

equivalently by the adjacency matrix.

The skeleton adjacency matrix is a symmetric ma-

trix with non-zero entries whenever two nodes are

connected by an edge and, thus, it codiﬁes an undi-

rected graph with the 1st node at the trachea. The

graph is traversed to obtain the binary tree that en-

codes airways anatomy. Each node is visited by iter-

atively traversing the graph starting from the 1st node

to create a list of all the visited nodes. This list cor-

respond to all parents the node is connected to and it

is used to remove edges from the adjacency matrix to

get the tree structure.

The ﬁnal binary tree is deﬁned by labelling each

branch according to its bronchial level and orienta-

tion (left, right) with respect the splitting branch at

the previous level. The tree top node corresponds

Towards a Videobronchoscopy Localization System from Airway Centre Tracking

353

Figure 1: Extraction of Lumen Centres using a Multi-scale MSER.

to the tracheal entry point and it is labelled ”0”. At

each new branch, the added nodes are labelled ”0”

or ”1” depending on the anatomical branch orienta-

tion (”0” for left, ”1” for right). To label branches

at each level, we use the orientation of the current

and children nodes with respect to the current node

parent. Let P

= (x

, y

, z

), P

= (x

, y

, z

) and

= (x

, y

, z

) be the 3D positions in the air-

way skeleton of the current node, n, its parent, p, and

current node children, ch

, i = 1, 2. Taking P

as ori-

gin of coordinates, such positions deﬁne a coordinate

change in R

given by the vectors

−→

= P

− P

−→

= P

− P

and

−→

= P

− P

. Each children

node n

is labelled ”0” or ”1” depending on the ori-

entation (clock wise or counter clock wise) of the ba-

sis given by

−→

. Since the orientation of

a coordinate change is determined by the sign of the

determinant of the matrix given by the vector coordi-

nates, children nodes are labelled according to:

Label(ch

) =



0 , if det((

−→

)) > 0

1 , otherwise

Label(ch

) = 1 − Label(ch

) (1)

for (

−→

) the matrix given by the column

vectors

−→

, i = 1, 2, 3.

2.2 Bronchial Path from Exploration

Videos

The computation of the navigated path has 2 stages:

extraction of lumen centres in images using a multi-

scale MSER over a likelihood map of lumen centre

location and tracking of MSER centres to endow con-

tinuity across the sequence.

Extraction of airway centres is based on likelihood

maps (S

anchez et al., 2014) with maximum values at

candidate centres. Under the assumption of a circular

luminal proﬁle in central navigation, such maps use a

single scale isotropic Gaussian kernel to characterize

dark image areas prone to belong to airway lumen.

Unlike (S

anchez et al., 2014), we compute several

maps using a bank of anisotropic Gaussian ﬁlters with

different orientations and scales to model non-circular

lateral bronchi and small distal levels. Gaussian ﬁlters

have been normalized by their L

norm to obtain more

uniform responses comparable across different scales

and degrees of anisotropy.

To compute local maxima consistent across scales

and orientations, we use maximal stable extremal re-

gions (MSER) in the multi-scale scheme sketched in

ﬁg.1. A set of MSER elliptical regions are extracted

for each likelihood map (ﬁg.1, step1). To endow

stability across likelihood maps and, thus, different

scales and orientations, MSER regions are ﬁrst split

according to their orientation using a histogram (ﬁg.1,

step2). For each orientation, elliptical MSER regions

are sorted by their area and grouped into N clusters

(ﬁg.1, step3). The union of all ellipses in each cluster

deﬁnes a set of N regions that are ﬁltered using an in-

clusion criteria. The overlapping between regions is

computed and those having maximum overlap are se-

lected. The collection of MSER centroids for all valid

regions are the set of lumen centres for each frame

(ﬁg.1, step4).

To endow centres with temporal continuity,

MSER candidates are tracked using a modiﬁed

Kalman ﬁlter (Haykin, 2004). For each lumen cen-

tre at a given frame, a Kalman ﬁlter (Haykin, 2004)

predicts its location in the next frame according to a

motion model (constant velocity, constant accelera-

tion). Such motion models are prone to fail in intra-

operative videos due to abrupt movements in scope

manual guiding and patient coughing. To reduce the

impact of these variations in motion model, we have

implemented a constant position tracker that uses a

state vector with zero velocity.

VISAPP 2017 - International Conference on Computer Vision Theory and Applications

354

2.3 CT-Navigation Matching

To match the navigation path to the binary tree en-

coding CT bronchial anatomy, we identify frames

traversing a higher bronchial level (binary tree level)

and orient the entering branches to chose the tree

node. A given frame at time, t, can be categorized

from the multiplicity of lumen centres, NLM, as:

1) Frame within same bronchial level (NLM

t+1

NLM

); 2) Frame approaching a bronchial level

(NLM

t+1

> NLM

); 3) Frame traversing a bronchial

level (NLM

t+1

< NLM

). Starting at the top node

of the binary tree, each time a frame traverses a

bronchial level, the tree level is increased and the path

node sequence is updated by adding ”1” or ”0” de-

pending on the entering branch orientation. The cen-

tre point with highest likelihood is considered to be

the scope current position and deﬁnes the entering

branch. Its orientation is deﬁned by its relative po-

sition with respect the disappearing centres. If the

x-coordinate is larger than the average x-coordinate

of the vanishing points, we consider that the node is

labelled ”1” and ”0” otherwise.

Figure 2 illustrates the main steps involved in our

CT-video matching based on the codiﬁcation of air-

ways anatomical landmarks. We show the skeleton of

a segmented CT scan (left image) that represents the

centre airway line and the ﬁnal binary tree data struc-

ture for the ﬁrst 3 bronchial levels (central image). We

have labelled the skeleton branching points according

to their corresponding binary tree nodes, so that the

green path would correspond to the node sequence

(1, 0, 1, 0). The most right images illustrate the iden-

tiﬁcation of lumen centres and the matching to the bi-

nary tree (central image) representing the exploration

bronchial path. Lumen centres in right images are

plotted in green, with dots indicating the one corre-

sponding to the scope current position. We show two

representative cases: a frame within same bronchial

level (top images) and a traversing frame (bottom im-

ages). The node sequence associated to these frames

is shown on the central tree in green.

3 EXPERIMENTS

3.1 Experiment1: Tracked Centres

Accuracy

We have compared under intervention conditions the

quality of centres tracked using Section 2.2 (la-

belled MSER) to the method in (S

anchez et al.,

2015b) exclusively based on local maxima (labelled

LMx). Methods have been applied to 3 ultrathin bron-

choscopy videos performed for the study of peripheral

pulmonary nodules at Hospital de Bellvitge. Videos

were acquired using an Olympus Exera III HD Ultra-

thin videobronchoscope. For each video, we consid-

ered one proximal (up to 6th division) and one dis-

tal (above 6th) fragments. The maximum bronchial

level achieved in our ultrathin explorations was within

10th and 12th, which is in the range of the maxi-

mum expected level reachable by ultrathin navigation

(Asano et al., 2013). Fragments included the most

common artefacts of intra-operative videos: broncho-

scope collision with the bronchial wall, bubbles due

to the anaesthesia and patient coughing.

For each fragment, we sampled 10 consecutive

frames every 50 frames. Such frames were annotated

by 2 clinical experts to set false detections and the po-

sition of missed centres. Inspired in crowd sourcing

strategies (Maier-Hein et al., 2015), annotations were

blended to get a unique ground truth using the inter-

section of the two annotated point sets as illustrated

in ﬁg.3. Ground truth sets were used to compute pre-

cision (Prec) and recall (Rec) for each set of consecu-

tive frames. These scores are taken for all such sets in

distal and proximal fragments for statistical analysis.

We have used a T-test for paired data to assess signiﬁ-

cant differences across methods average precision and

recall and conﬁdence intervals, CI, to report average

expected ranges. Tests and CIs have been computed

at signiﬁcant level α = 0.05.

Table 1 reports CIs for each score and method at

proximal and distal levels, as well as, p-values for the

difference between MSER and LMx scores. At prox-

imal level, both methods perform equally, but MSER

keeps its quality scores at distal levels. This intro-

duces signiﬁcant differences (p-val< 0.05) in dis-

tal and total Prec and Rec. Such differences are

larger for Rec, with a CI for the difference equal to

[0.05, 0.16] for distal bronchi and [0.03, 0.16] overall.

It is worth noticing that the proposed method always

has a 100% of precision and a recall over 86%, with

non-signiﬁcant differences between distal and proxi-

mal levels (p-val > 0.7).

To validate the stability of our tracked centres in

full explorations, we have applied our MSER track-

ing to one of the complete videos. The chosen video

starts at carina, reaches the 11th level and includes

back and forth navigation with bronchoscope rota-

tion. Concerning image quality, there are saturation

illumination artefacts at most distal levels and some

fragments were recorded using narrow band imaging.

The original video with the tracked centres on each

image frame with a colour legend indicating candi-

date centres discarded by the Kalman ﬁlter, tracked

Towards a Videobronchoscopy Localization System from Airway Centre Tracking

355

Figure 2: CT-Video Anatomical Matching. Codiﬁcation of the bronchial anatomy from CT branch points (right image) and

binary tree coding for identiﬁcation of the navigated path inside the bronchial tree (central and left images).

Figure 3: Manual Annotation of Tracked Centres.

Table 1: Average precision and recall conﬁdence intervals for centre tracking.

Proximal Distal

MSER LMx p-val MSER LMx p-val

Prec [1.0, 1.0] [0.95, 1.0] 0.07 [1.0, 1.0] [0.93, 0.99] 0.02

Rec [0.86, 1.0] [0.81, 0.96] 0.4 [0.82, 0.97] [0.74, 0.87] 7e − 4

centres and scope position is available at https://

www.youtube.com/watch?v=h8kUrzDjSKo. Figure

4 shows representative frames at proximal (a,b) and

distal (c,d) levels.

3.2 Experiment2: Anatomical Matching

Accuracy

We have tested our methodology in virtual explo-

rations of CT volumes from 3 patients coming from

Hospital de Bellvitge. CT scans were acquired with

a 320-detector row Aquilion ONE, Toshiba CT scan

and 0.5 mm thickness. Each CT volume was man-

ually segmented using AMIRA to create a triangular

mesh in .obj format for navigation simulation. Virtual

explorations were generated using a simulation plat-

form developed in Unity (Borras et al., 2015), which

allows the modelling of the scope camera and inter-

active point of view navigation. For this experiment,

the bronchoscope was simulated in central navigation

without rotation around the scope long axis (see the

Discussion Section 4 for clariﬁcations about this as-

sumption). The camera position inside the bronchial

VISAPP 2017 - International Conference on Computer Vision Theory and Applications

356

Figure 4: Representative frames at proximal, (a,b) and distal levels, (c,d).

tree was exported to deﬁne the codiﬁcation of the

ground truth path in the binary tree.

For each CT scan, 15 virtual explorations with

different branching paths and bronchial levels (from

4th to 12th) were deﬁned. For each path, the binary

tree node coding for the ground truth camera posi-

tion was compared to the code extracted from virtual

videos. Node path coding was compared in terms of

True Positives Nodes (TPN) and True Path Represen-

tations (TPR). For a given exploration, a node is con-

sidered to be a TPN if its label coincides with the GT

node label. The number of consecutive TPN achieved

from the 1st node divided by the path node length de-

ﬁnes TPR. As before, we have used a T-test to detect

signiﬁcant differences across proximal and distal lev-

els and CIs at signiﬁcance α = 0.05 to report average

TPN and TPR.

Table 2 reports CIs for average TPN, TPR per-

centages at proximal and distal levels and p-values

of the T-test for the difference. The detection rate of

Towards a Videobronchoscopy Localization System from Airway Centre Tracking

357

Table 2: Comparison of TPN and TPR at proximal and distal levels.

Proximal Distal p-val

TPN [70.32%, 85.86%] [67.32%, 81.18%] 0.4397

TPR [72.16%, 90.71%] [63.91%, 79.53%] 0.0370

nodes along the whole path is not signiﬁcantly dif-

ferent at distal and proximal levels, with a CI for

the difference equal to [−6.1%, 13.8%]. Concern-

ing the percentage of correct paths reached from the

trachea, there are signiﬁcant differences with a drop

between [1.3%, 15.9%]. This drop in performance

mainly arises from false distal centres due to illumi-

nation artefacts in simulated data and a suboptimal la-

belling of CTs (see Discussion Section 4 for details).

4 DISCUSSION AND

CONCLUSIONS

We have introduced a localization system for lung

cancer biopsy guidance based on identiﬁcation of air-

ways centres. The system bases on a codiﬁcation

of airways using a binary tree and an airway lumen

extraction from videobronchoscopy with an ultrathin

bronchoscope. Our experiments validate centre ex-

traction in 3 intra-operative ultrathin videos and the

CT-video path retrieval in simulated data.

Results in ultrathin bronchoscopy videos indicate

high equal performance of our centre tracker based

on maximal stable regions at proximal and distal

levels. In particular, there are not any false detec-

tions (Prec=1) and the rate of missed centres is un-

der 15% (Rec> 0.85). The decrease in recall is due

to the model assumptions used to compute the like-

lihood maps (S

anchez et al., 2014), which even us-

ing anisotropic Gaussians favour circular geometries.

This could be solved by considering the bronchial

structure of each patient extracted from the CT-scan.

This expected local conﬁguration of airway centres

could be used during the tracking step to retrieve

missing lateral bronchi. Still, results are promis-

ing enough (see the full exploration at https://

www.youtube.com/watch?v=h8kUrzDjSKo) to en-

courage a use in a navigation system to explore the

beneﬁts of using video information for biopsy guid-

ance to peripheral pulmonary nodules.

The potential of our centre tracking for CT-

navigation matching is illustrated in the experiment

with simulated data. It is worth noticing that the qual-

ity of tracked centres is enough to correctly match

80% of 6th level paths and 70% of 12th level paths us-

ing only multiplicity of the tracked centres. Of course

this matching criteria is not powerful enough to cope

with artefacts in centres detection, which introduce

false nodes in the tree path coding for distal paths. We

think that codiﬁcation of the full MSER regions hier-

archy would provide a more solid representation of

bronchial branching in images and,thus, increase the

level successfully matched. Aside, the labelling of the

CT tree should take into account the local orientation

(Frenet basis) of the path traversing bronchial level

to obtain a labelling consistent with image projection

orientation. Finally, simulations were done assuming

central non-rotational navigation. This is not a critical

assumption, provided that the goal was to validate a

CT-video matching based on lumen anatomical land-

marks as alternative to image registration. However,

to have a system operative in clinical conditions, the

criteria used to label image centre lumens should be

modiﬁed to allow rotations of the bronchoscope. This

could be achieved by computing the intra-frame rota-

tion to adapt the position of centres in central naviga-

tion to the current exploration and it is future work.

ACKNOWLEDGMENTS

Work supported by projects DPI2015-65286-R, FIS-

ETES PI09/90917, 2014-SGR-1470 and Fundaci

Marat

o TV3 20133510. Also supported by CERCA

Programme / Generalitat de Catalunya. Finally, Deb-

ora Gil is supported by the Serra Hunter Program.

REFERENCES

Aberle, D., Adams, A., and et. al. (2011). Reduced

lung-cancer mortality with low-dose computed tomo-

graphic screening. N Engl J Med, 365:395–409.

Asano, F., Shinagawa, N., and et. al. (2013). Virtual

bronchoscopic navigation combined with ultrathin

bronchoscopy. arandomized clinical trial. AJRCCM,

188(3):327–333.

Borras, A., Gil, D., and et al. (2015). A virtual broncho-

scopic tool to explore the impact of physical restric-

tionsin bronchoscopy planning. In MICCAI IMIM.

Donnelly, E. and Edwin, F. (2012). Technical parameters

and interpretive issues in screening computed tomog-

raphyscans for lung cancer. JTI, 27(4):224–229.

Eberhardt, R., Kahn, N., and et. al. (2010). Lungpointa

new approach to peripheral lesions. JTO, 5(10):1559–

1563.

VISAPP 2017 - International Conference on Computer Vision Theory and Applications

358

Gildea, T., Mazzone, P., and et. al. (2006). Electromag-

netic navigation diagnostic bronchoscopy: a prospec-

tive study. AJRCCM, 174(9):982–989.

Haykin, S. (2004). Kalman ﬁltering and neural networks,

volume 47. John Wiley & Sons.

Hofstad, E., Sorger, H., and et. at (2015). Automatic regis-

tration of ct images to patient during bronchoscopy- a

clinicalpilot study. In ECBIP.

Kerschnitzki, K. and et al. (2013). Architecture of the os-

teocyte network correlates with bone material quality.

JBMR, 28(8):1837–45.

o, X., Feuerstein, M., and et al (2012). Development

and comparison of new hybrid motion tracking for

bronchoscopicnavigation. Medical image analysis,

16(3):577–596.

Maier-Hein, L., Kondermann, D., and et. al. (2015).

Crowdtruth validation: a new paradigm for validat-

ing algorithms that relyon image correspondences. IJ-

CARS, 10(8):1201–12.

Manhire, A., Charig, M., and et. al. (2003). Guide-

lines for radiologically guided lung biopsy. Thorax,

58(11):920.

Matas, J., Chum, O., Urban, M., and Pajdla, T. (2004). Ro-

bust wide-baseline stereo from maximally stable ex-

tremal regions. IMAVIS, 22(10):761–767.

Mirota, D., Ishii, M., and Hager, G. (2011). Vision-based

navigation in image-guided interventions. Annual re-

view of biomedical engineering, 13:297–319.

Reynisson, P., Leira, H., and et. al (2014). Navigated bron-

choscopy: a technical review. JBIP, 21(3):242–264.

anchez, C., Bernal, J., and et al (2014). On-line lumen

centre detection in gastrointestinal and respiratory en-

doscopy. In MICCAI-CLIP, volume 8361 of LNCS,

pages 31–38.

anchez, C., Bernal, J., S

anchez, F. J., Diez, Marta an-

dRosell, A., and Gil, D. (2015a). Toward online

quantiﬁcation of tracheal stenosis from videobron-

choscopy. IJCARS, 10(6):935–945.

anchez, C., Gil, D., and et al (2015b). Navigation path

retrieval from videobronchoscopy using bronchial

branches. In CLIP, LNCS 9401.

Van Uitert, R. and Bitter, I. (2007). Subvoxel precise skele-

tons of volumetric data based on fast marching meth-

ods. Medical physics, 34(2):627–638.

Towards a Videobronchoscopy Localization System from Airway Centre Tracking

359