Automatic Detection of Air Holes Inside the Esophagus in CT Images.
ABSTRACT Air holes inside the esophagus can be used to localize the esophagus in computed tomographic (CT) images. In this work we pre- sent a technique to automatically detect esophageal air holes in this modality. Our technique is based on the extraction of a volume of interest, air segmentation by thresholding and classification of respiratory and esophageal air using a priori knowledge about the connectivity of air voxels. A post-processing step rejects wrong results from artifacts in the CT image. We successfully tested our algorithm with clinical data and compared the detection results of a human expert and our technique.
- SourceAvailable from: Alexander Cavallaro[Show abstract] [Hide abstract]
ABSTRACT: Automatic segmentation of the esophagus from CT data is a challenging problem. Its wall consists of muscle tissue, which has low contrast in CT. Sometimes it is filled with air or remains of orally given contrast agent. While air holes are a clear hint to a human when searching for the esophagus, we found that they are rather distracting to discriminative models of the appearance because of their similarity to the trachea and to lung tissue. However, air inside the respiratory organs can be segmented easily. In this paper, we propose to combine a model based segmentation algorithm of the esophagus with a spatial probability map generated from detected air. Threefold cross-validation on 144 datasets showed that this probability map, combined with a technique that puts more focus on hard cases, increases accuracy by 22%. In contrast to prior work, our method is not only automatic on a manually selected region of interest, but on a whole thoracic CT scan, while our mean segmentation error of 1.80mm is even better.Medical image computing and computer-assisted intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention. 01/2010; 13(Pt 1):95-102.
- [Show abstract] [Hide abstract]
ABSTRACT: Automatic segmentation of the esophagus from CT data is a challenging problem. Its wall consists of muscle tissue, which has low contrast in CT. Sometimes it is filled with air or remains of orally given contrast agent. While air holes are a clear hint to a human when searching for the esophagus, we found that they are rather distracting to discriminative models of the appearance because of their similarity to the trachea and to lung tissue. However, air inside the respiratory organs can be segmented easily. In this paper, we propose to combine a model based segmentation algorithm of the esophagus with a spatial probability map generated from detected air. Threefold cross-validation on 144 datasets showed that this probability map, combined with a technique that puts more focus on hard cases, increases accuracy by 22%. In contrast to prior work, our method is not only automatic on a manually selected region of interest, but on a whole thoracic CT scan, while our mean segmentation error of 1.80mm is even better. Additional material can be found at http://www5.informatik.uni-erlangen.de/fileadmin/Persons/FeulnerJohannes/RoiWithoutAndWithSegmentation.avi .09/2010: pages 95-102;
- [Show abstract] [Hide abstract]
ABSTRACT: Being able to segment the esophagus without user interaction from 3-D CT data is of high value to radiologists during oncological examinations of the mediastinum. The segmentation can serve as a guideline and prevent confusion with pathological tissue. However, limited contrast to surrounding structures and versatile shape and appearance make segmentation a challenging problem. This paper presents a multistep method. First, a detector that is trained to learn a discriminative model of the appearance is combined with an explicit model of the distribution of respiratory and esophageal air. In the next step, prior shape knowledge is incorporated using a Markov chain model. We follow a “detect and connect” approach to obtain the maximum a posteriori estimate of the approximate esophagus shape from hypothesis about the esophagus contour in axial image slices. Finally, the surface of this approximation is nonrigidly deformed to better fit the boundary of the organ. The method is compared to an alternative approach that uses a particle filter instead of a Markov chain to infer the approximate esophagus shape, to the performance of a human observer and also to state of the art methods, which are all semiautomatic. Cross-validation on 144 CT scans showed that the Markov chain based approach clearly outperforms the particle filter. It segments the esophagus with a mean error of 1.80 mm in less than 16 s on a standard PC. This is only 1 mm above the interobserver variability and can compete with the results of previously published semiautomatic methods.IEEE Transactions on Medical Imaging 07/2011; · 3.80 Impact Factor
Automatic Detection of Air Holes
inside the Esophagus in CT Images
Andreas Fieselmann1,2, Stefan Lautenschl¨ ager2, Frank Deinzer2, Bj¨ orn Poppe1
1Medical Radiation Physics, University of Oldenburg, Oldenburg, Germany
2Siemens AG Medical Solutions, Forchheim, Germany
Abstract. Air holes inside the esophagus can be used to localize the
esophagus in computed tomographic (CT) images. In this work we pre-
sent a technique to automatically detect esophageal air holes in this
modality. Our technique is based on the extraction of a volume of interest,
air segmentation by thresholding and classification of respiratory and
esophageal air using a priori knowledge about the connectivity of air
voxels. A post-processing step rejects wrong results from artifacts in the
CT image. We successfully tested our algorithm with clinical data and
compared the detection results of a human expert and our technique.
The esophagus is a muscular tube with a diameter of approximately 10 mm.
It passes from the pharynx through the mediastinum into the stomach and has
direct contact with the left atrial posterior wall whereas the exact route varies
individually . It is important to know the position of the esophagus during
atrial ablation therapy. This intervention treats atrial fibrillation by radiofre-
quency catheter ablation of the pulmonary veins inside the left atrium. If the
power of the catheter at the contact area of the left atrium and the esophagus is
too high, thermal injury of the esophagus may occur and a fistula that connects
the esophagus and the left atrium can develop . This situation is potentially
fatal because air is able enter into the left atrium.
A segmentation of the esophagus is helpful to plan the ablation strategy and
decrease the power at its contact area with the left atrium. Unfortunately, the
contrast of the esophagus in medical images (cardiac CT, cardiac C-arm CT 
or cardiac MRI) is low which makes edge-based segmentation difficult. A feature
to identify the position of the esophagus are small air holes inside the esophagus
(Fig. 1). They are a normal phenomenon and are visible in most CT images,
although their number and sizes are highly variable.
Very few studies investigated the segmentation of the esophagus in tomo-
graphic images [4, 5, 6]. Rousson et al.  detect esophageal air holes to refine
their segmentation result. After their segmentation algorithm has estimated a
centerline of the esophagus low intensity voxels are detected in the neighborhood
of the centerline. The complete air holes are obtained by means of a region grow-
ing algorithm. Two points on the centerline and a segmentation of the aorta and
left atrium are needed as input to the algorithm. The two other studies [5, 6] use
optic flow and deformable models respectively to segment the esophagus but do
not detect esophageal air holes. Our technique detects esophageal air holes fully
automatic without any user input and can be used to initialize an algorithm to
segment the esophagus.
2Materials and Methods
Our algorithm consists of three main steps which are described in the following
sections and illustrated in Fig. 2.
2.1Automated Extraction of a Volume of Interest
The algorithm confines all computations to a volume of interest (VOI) in order to
decrease computational time and the number of wrong results in the subsequent
analysis. It is extracted relative to a reference point p = (px,py,pz)Tat the left
atrium which is computed by projecting the center of mass of the left atrium
onto its posterior wall. A segmentation of the left atrium is available because it
is the primary organ of interest and always segmented before the esophagus in
the clinical workflow.
We use a 2D rectangular region of interest centered at (px+tx,py+ty)Tin all
axial slices. The side lengths sx, syand t = (tx,ty)Tare determined from a set
of CT images such that the esophagus is always inside the VOI. Our parameters
measured in mm and with orientation as shown in Fig. 1 are t = (−15,−20)T,
sx= 100 and sy= 100.
2.2Classification of Esophageal Air Holes
Most visible air in a cardiac CT scan exists inside the respiratory organs: the
lungs, the bronchi and the trachea. This respiratory air constitutes one connected
Fig.1. Cardiac CT images with esophageal air holes: (a) axial and (b) sagittal view
component. Another location in the mediastinum where air can exist is inside
Air has a very low Hounsfield unit (HU) value in CT images. In our algorithm
the entire air is segmented by applying a threshold value Iairat -400 HU. This
threshold value has been suggested by Kemerink et al.  to segment the lung
parenchyma. All voxels with intensities below -400 HU are labeled as air.
In the next step the air is classified into respiratory or esophageal air. The
respiratory air is classified using a 3D volume growing algorithm  whose seeds
are placed at the left and right boundaries of the VOI at voxels that have been
labeled as air. The output of the volume growing algorithm is subtracted from
the initial air image. In the resulting image all air voxels not connected to the
boundaries are still labeled. These voxels are classified as esophageal air. Finally,
esophageal air holes are found by identifying connected components of esophageal
Artifacts in the CT image with low HU values can be misclassified as esophageal
air. An important example are dark streaks due to beam hardening of the
X-rays . They occur very close to high intensity areas. These high intensity
areas originate e.g. from the use of a contrast agent. Post-processing is needed
to recognize and reject artifacts misclassified as esophageal air.
Our algorithm recognizes beam hardening artifacts by computing the image
intensity histogram in a certain region around each esophageal air hole. If the
histogram contains intensities greater than 600 HU then the air hole is rejected.
The histogram region is determined by dilating the bounding box of the air hole
isotropically by 5 mm.
We validate our algorithm with 7 pre-interventional contrast-enhanced cardiac
CT scans of patients treated with atrial ablation therapy. The scans were ac-
Fig.2. Flow chart of the algorithm to detect esophageal air holes in a CT image
Table 1. Esophageal air holes detected by a human expert and the algorithm in 7
patient scans. Caudal and cranial are defined relative to the left atrial center
number at caudal/cranial side
number at caudal/cranial side
misclassified esoph. air holes
number of rejected artifacts
minimum size in voxels
maximum size in voxels
3/911/115/2 10/4 6/3 4/52/1
quired using a 16-slice MDCT scanner (SOMATOMR
Medical Solutions) with an in-plane resolution of approximately 0.35·0.35 mm2
and a slice thickness of approximately 0.6 mm. A human expert labels all esoph-
ageal air holes manually and the left atrium is segmented using the software
rithm in C++ using the open-source software library Insight Toolkit .
Our algorithm detects esophageal air holes in all 7 patient scans (Table 1)
but misses air holes that have intensities greater than Iairwhich are found by
the human expert. The number of misclassified air holes is very low and all
beam hardening artifacts are successfully recognized and rejected. The sizes of
detected air holes vary strongly. The minimum size is one pixel volume but this
does not correspond to the true size of the air hole which contains some voxel
intensities greater than Iair. The majority of air holes are found caudal to the
left atrial center before the esophagus enters into the stomach. Fig. 3 shows air
holes cranial (a-c) and caudal (d-f) to the left atrial center. The computational
time is about 5 seconds for one patient scan.
?Sensation 16, Siemens AG
?InSpace EP (Siemens AG Medical Solutions). We implement our algo-
Our algorithm successfully detects esophageal air holes in cardiac CT scans and
requires no input from the user. It makes use of the connectivity of respiratory
air voxels which are not connected to esophageal air voxels. A possible extension
of our algorithm is to use an adaptive threshold to segment the air. Because only
a few air holes are sufficient to localize the esophagus it is not a high priority to
detect all air holes.
The locations of air holes reveal the position of the esophageal lumen and the
contour of the esophagus can be estimated using this knowledge. It is possible
to visualize the air holes directly or use the air holes to initialize an algorithm
to segment the esophagus.
Fig.3. Regions of cardiac CT images with labeled esophageal air holes: (a-c) cranial
of the atrial center, (d-f) caudal of the atrial center
1. Tsao HM, Wu MH, Higa S, et al.
and left atrium: Implication for catheter ablation of atrial fibrillation.
2. Pappone C, Oral H, Santinelli V, et al.
cation of percutaneous transcatheter ablation of atrial fibrillation. Circulation.
3. Lauritsch G, Boese J, Wigstr¨ om L, et al. Towards cardiac C-arm computed to-
mography. IEEE Trans Med Imaging. 2006;25(7):922–34.
4. Rousson M, Bai Y, Xu C, et al. Probabilistic minimal path for automated esoph-
agus segmentation. Proc SPIE. 2006;6144:49–4H.
5. Huang TC, Zhang G, Guerrero T, et al. Semi-automated CT segmentation us-
ing optic flow and Fourier interpolation techniques. Comput Methods Programs
6. Ragan D, Starkschall G, McNutt T, et al.
computed tomography segmentation using deformable models.
7. Kemerink GJ, Lamers RJS, Pellis BJ, et al. On segmentation of lung parenchyma
in quantitative computed tomography of the lung. Med Phys. 1998;25(12):2432–9.
8. Handels H. Medizinische Bildverarbeitung. Stuttgart: Teubner; 2000.
9. Kalender WA. Computed Tomography.
10. Ibanez L, Schroeder W, Ng L, et al. The ITK Software Guide. 2nd ed. Kitware
Anatomic relationship of the esophagus
Atrio-esophageal fistula as a compli-
2nd ed. Erlangen: Publicis Corporate