Bronchial Tree Analysis


Software Development for the Detection and Assessment of Small Airways Disease in COPD With Multi Slice Computed Tomography

Netherlands Asthma Foundation EBO 2001

Maarten Nieber, Berend Stoel

Background

Figure 1Pathological characteristics found in small airways with Chronic Obstructive Pulmonary Disease (COPD) are considered to play an important role in the progression of COPD and the chronicity of the disease. This new interest into the progression characteristics of COPD has put the pressure on the development of non-invasive markers reflecting pathological changes recently identified in small airways measured in surgically resected lung specimen. One approach may be the utilization of the image data obtained with Computed Tomography (CT) of the chest. The quantitative assessment of small airways by computed tomography is a relatively new research area, since its progress is strongly determined by the technological progress in the field of Computed Tomography. After the introduction of High Resolution CT (HRCT) in the early 1980's, CT images of the bronchi were primarily assessed by radiologists in a qualitative manner. With simple measuring tools on the CT-console for determining distances between two points in the image, quantitative assessments of bronchi can be made. These measurements are, however, strongly degraded by a considerable inter- and intra-observer variability. Moreover, no absolute CT criteria of normal bronchial diameters have been determined thus far and therefore a diagnosis based on these measurements remains somewhat subjective. Since the scan time of one image by these HRCT scanners was approximately 1 second, only a part of the lungs could be scanned in one breath hold. Therefore, airways were assessed in separate 2-dimensional cross-sections of the lungs.

The parameters, which were measured in these cross-sections, are:

  1. Area of the lumen;
  2. Absolute wall thickness;
  3. Wall thickness related to the outer diameter of the bronchus (WA%);
  4. Circumference of the internal boundary of the bronchus.
  5. Internal diameter of the bronchus divided by the diameter of the adjacent pulmonary artery: bronchoarterial ratio (B/A).

The drawback of the B/A parameter is that an increase in B/A may actually reflect a reduction in the pulmonary artery size, due to a reduction in lung ventilation. Furthermore, Lynch demonstrated that most of a group of 27 normal subjects showed bronchi with an internal diameter exceeding that of the adjacent pulmonary artery branch. Therefore, the bronchial wall thickness, related to the outer diameter, (WA%) should be considered as well. Possibly, wall irregularity measures may also be informative measures.

All these measurements were made using only a limited number of 2-dimensional (HRCT) images of the lungs. For the assessment of the entire bronchial tree, single Spiral CT and even Electron Beam CT would be of limited use, due to its inadequate resolution in the z-direction (perpendicular to the image plane).
Despite these limitations, some techniques have been developed which automatically detect the bronchial tree in 3-dimensional CT-data. These algorithms can detect the bronchial tree up to the fourth generation.
The recent introduction of the multi-detector technique in CT offers the opportunity of an improved resolution in the z-direction, while still covering an adequate volume. This will make it possible to detect the bronchial tree up to higher generations, which will facilitate a more accurate and more comprehensive assessment of the bronchial tree, described in this grant application.
Theoretically, MSCT scanners allow the detection of bronchi beyond the level of the sub-segmental bronchi, up to a level between the 5th and 10th generation (3.5 - 1.3 mm in diameter, see table 1). These dimensions are well within the resolution of a MSCT scanner (in-plane: ± 0.28mm; slice thickness: ± 0.5mm), but influences of blurring due to (cardiac) motion are unknown.

Generation

Name

Diameter [mm]

Length [mm]

0

Trachea

18.0

120.0

1

Primary bronchi

12.2

47.6

2

Lobar bronchi

8.3

19.0

3

Segmental bronchi

5.6

7.6

4

Subsegmental bronchi

4.5

12.7

5-10

Small bronchi

3.5-1.3

10.7-4.6

11-13

Bronchioli

1.09-0.8

3.9-2.7

Table 1. Average dimensions of the bronchial tree.

Goals

General aim of the project:

  • To detect and visualize distal airways of the bronchial tree with Multi Slice Computer Tomography of the chest.
  • To assess the differences in airway wall thickness or diameter between subjects with healthy lungs and in patients with COPD.

Software-specific research questions:

  • Up to which generation can the internal boundaries of the bronchial tree be detected? The aim is to detect bronchi beyond the level of the sub-segmental bronchi, up to a level between the 5th and 10th generation, with diameters between 3.5 mm and 1.3 mm.
  • Can the surrounding arteries be detected at a level between the 5th and 10th generation? And can this artery detection prevent false positive detection of the bronchi, effectively.
  • Can the outer boundaries of the bronchial tree be detected at the same level?

 Clinical-related research questions:

  • Which parameters that can be calculated from these inner and outer boundaries are clinically relevant for assessing small airway disease?
  • Can these measurements discriminate different subgroups within the group of COPD patients?
  • Are the parameters possible targets for pharmacologic intervention? .

Approach

The analysis consists of the following steps: 

  1. Bronchial Tree Detection & Partitioning into branches (Figure 1)
    The bronchial tree is detected using wave propagation. The algorithm requires the user to place a seed point in the trachea. Using wave propagation, new points surrounding the seed point that match the density of air are added to the bronchial tree. The resulting segmented bronchial tree can be adjusted manually with dedicated tools for automatically adding or removing parts of the segmentation. Subsequently, the detected bronchial tree is automatically divided into branches that correspond to the generations in the bronchial tree.
  2. Selection of a branch and calculation of a reformatted slice (Figure 2)
    When the user selects a branch, the software shows the bronchial major axis and a plane perpendicular to the selected branch, which is positioned at the central location of the branch. By moving the slice plane along the bronchial main axis, the user selects the location where a reformatted 2D cross-sectional image of the bronchus is computed. The central position is suitable for measuring the area of the bronchial lumen. For measuring
  3. Contour detection in the reformatted slice (Figure 3)
    The inner wall contour (around the bronchial lumen) and outer wall contour (around the outside of the bronchial wall) are identified in the cross-sectional image. The result of the inner and outer contour detection is overlaid on the original image. In those cases where only a part of the wall-to-parenchyma transition is visible in the image, the outer wall contour is truncated automatically.
  4. Calculation of bronchial geometry parameters
    The following quantities are automatically calculated from the resulting contours:
    • Cross-sectional area of the lumen 
    • Estimated total bronchus area, including the wall (Ao)
    • Estimated wall area (WA)
    • Percentage wall area (WA relative to Ao)
    • Average wall thickness
    • Circularity of the bronchus

Status

Finished

Gallery

Figure 1
Figure 1. Detection and labeling of bronchial tree.

Figure 2
Figure 2. Selection of a bronchus and calculation of perpendicularly reformatted slice .

Figure 3
Figure 3. Contour detection of bronchial lumen and wall.

Contact

For further information, please contact:
B.C. Stoel, PhD.
Division of Image Processing
Department of Radiology, 1-C2S
Leiden University Medical Center
P.O. Box 9600
2300 RC Leiden
The Netherlands
Tel. +31 (0)71 526 1911
Fax. +31 (0)71 526 6801
e-mail: B.C.Stoel@lumc.nl