with reference to the embodiments described hereinafter as illustrated by the following Figures.
The human lung consists of two major parts, the left lung and the right lung. There are three lobes in the right lung, which are separated by the so-called major fissure and minor fissure. The left lung shows a slightly different structure. Because there is no defined minor fissure, it consists of only two lobes, whereby the part that anatomically corresponds to the right middle lobe is merged with the upper lobe. Each lobe is again divided into two or more lung segments of which ten exist for each side of the lung. These segments are supplied by a complex system of branching trees that conduct blood and air into the distal regions where the gas exchange takes place. The bronchial tree has a pipe structure that is filled with air. It starts at the trachea and extends into the distal regions repeatedly splitting into smaller and smaller branches. In the human lung, the splitting occurs usually in bifurcations, e.g. the parent branch splits up into two child branches, but trifurcations also exist. The general tendency for child branches is that they decrease in diameter and length although this might be different in individual cases. Siblings don't necessarily have the same diameter. The bronchi are classified into lobar bronchi that supply the lobes, segmental bronchi, that supplying the individual segments, and sub-segmental bronchi. The bronchial wall surrounds the air-filled lumen of the bronchi. The thickness of this wall is correlated to the diameter of the segment in the sense that it gets thinner for smaller diameters. High-resolution multi-slice CT reveals bronchi segments in the 6th branching generation and higher which have diameters in the mm range. For diagnosis and treatment of asthmatic and emphysematic patients, the bronchial lumen, bronchial wall thickness, and the ratio of inner bronchial to accompanying arterial diameter are parameters which are used in clinical practice in order to detect and quantify airway narrowing, bronchial dilation, bronchial wall thickening, bronchiectasis, hyperresponsiveness, etc.
Within the first step 100, the tracheobronchial tree is automatically segmented. The segmentation starts with loading a three-dimensional image set (3D-image set) of a thorax. The 3D-image set is preferably acquired with a high resolution CT scanner, such as a multi-array CT scanner. The high resolution refers to a slice thickness of about 1.0-1.3 mm or less. Other 3D-image sets that are acquired by a scanner that can produce such a high resolution image set, for example an MR scanner, a 3D-RA scanner, PET scanner, or SPECT scanner, etc. can be used too. Further the 3-D image set can be acquired with and without contrast agent, cardiac or respiratory gating. If a CT 3D-image set is used, the lung and trachea area can be segmented out of the overall 3D-image set of the thorax by setting a Hounsfield threshold, f.e. at −500 HU and identifying all 3D-connected voxels below the Hounsfield threshold. Then, the lung and trachea area is identified as the largest component of 3D-connected voxels that is not touching the image boundaries. Next, the trachea must be determined. For this purpose, the first voxel in a direction perpendicular to the plane of the slices, i.e the z-direction, that belongs to the lung and trachea area is found, and also the last voxel in this direction, since the scan-direction can be head-to-feet or feet-to-head. Of these two voxel positions, the one is chosen which is more central in the plane of the slices, i.e. the xy-direction. If the image set comprises descriptive data indicating the scanning direction, this descriptive data can be used to determine the trachea.
Within the next step 102, the centerlines of the trachea, segmental bronchi and smaller airways, i.e. the sub-segmental bronchi are extracted. Further the branching points of the tree structure are determined. This step is based on a front propagation approach which detects “leakages” into the parenchymal tissue, see also T. Schlathölter, C. Lorenz, I. C. Carlsen, S. Renisch, T. Deschamps, Simultaneous Segmentation and Tree Reconstruction of the Airways for Virtual Bronchoscopy.
Proceedings SPIE Medical Imaging 2002, SPIE vol. 4684, part 1, pp. 103-113. Here, the front propagation method is used in conjunction with an anatomical model of the tracheobronchial tree. The front propagation method is a type of region growing technique that uses a concept motivated from physical wave-front propagation and that is based on the physical principle of least action. The front propagation method uses a fast marching algorithm, for example as described in T. Deschamps, L. D. Cohen, Minimal Paths in 3D images and application to virtual endoscopy, Lecture Notes in Computer Science: Computer Vision—ECCV 2000; 1843:543-557.
The front propagation equation used is of the type:
|∇T|F=1 (1)
where F(x) is the speed function of the front and T(x) denotes the time value when the front reaches the point x. A stepwise constant speed function is used of the following form:
with t being a threshold value just above the bronchial lumen, and I(x) denoting a gray value at the point x.
The front propagation method keeps a list of branches that have to be grown. This list is initialized with the trachea. After initialization the algorithm loops over a sequence of growing, branch detection, and branch validation.
Growing: consecutively, one branch is taken from the list and is grown according to the modified fast marching algorithm described above. Each branch keeps a reference to its initial radius (ri) and compares this after every grow step to the actual radius. When the current branch approaches a bifurcation, the actual radius increases and finally exceeds the initial radius times a multiplication factor α (e.g. α=1.1).
Branch detection: when the actual radius exceeds (α*ri, a check for branching is performed. Using α, the execution of the computationally expensive connectivity checking process can be reduced. In case no branching is detected, α is increased about 0.1 and the grow process is continued. In the case of branching, the validity of the current branch is checked. This process is responsible for the detection of leakage. When a branch is detected, this is stored for example in a linked list structure that represents the branching points of the tree structure.
Branch validation: After branching occurred, the validity of the parent branch B of the branches Bi can be verified. Validation is responsible for rejecting branches that most probably represent leaked regions. Two criteria: radius and connectivity are used for the validation.
Radius: Since the grid point distribution of each branch is known from the segmentation result, it is possible to calculate its covariance matrix. Using a cylindrical model of the tracheobronchial tree, it is possible to estimate an average radius of the branch using the lowest two eigenvalues (EV) of the covariance matrix:
Since generally the radius is decreasing with increasing branch order a radius smaller than β*rmin (with rmin being the smallest radius of all ancestors) indicates leakage. β is chosen to be greater than 1 to provide a safety margin to the internal variability of the radius of the branches.
Connectivity: By checking the neighbor voxels of all surface voxels of a branch B, one can find the number of branches, which are in the direct vicinity of B. If one compares the number of different branches in the direct neighborhood with the maximum number of allowed branches (γ) one can detect leakages. γ should be set to an integer number greater than three. Three neighbors is the usual case since a branch usually has a parent and two children. Three or more children are also possible, thus this parameter should be chosen carefully, not too low and not too large (e.g. γ=5).
For valid branches, the unconnected regions of the front are used to initialize new branches, which are stored in the branch list; for invalid branches they are discarded. Thus during the growth, each “front voxel” belongs to one of several 3D-connected growth fronts. If one of these fronts becomes too large, then it is considered “leakage” and this front is frozen, and only voxels from other fronts are propagated.
The region growing of the tracheobronchial tree can be repeated several times, starting with a high Hounsfield thresholds (e.g. −800 HU), and then descending to lower thresholds (e.g. down to −900 HU in steps of 20 HU), where the resulting voxels from each iteration are taken as seeds for the next iteration.
The centerlines can be determined by computing a distance map for the segmented volume of the bronchial tree, giving the distance for each voxel to the nearest non-bronchi voxel. The distance can be derived from the radius. Such a non-bronchi voxel is part of the surrounding lung parenchyma tissue. All bronchial centerlines can be written into a table with the original trachea seed-point as the endpoint and the most distal point as the start-point. Then all bronchi can be traced for left and right lung separately, and measurements of the clinical parameters below can be taken at each trace point.
Within the final step 104, clinical parameters are determined for the segmented tracheobronchial tree and these are displayed preferably together with the segmented tree. At each point along all bronchial centerlines, the bronchial lumen that is equal to two times the radius of the inner bronchial wall, the radius to the outer bronchial wall, and the thickness of the accompanying artery is measured. Thus, the mean wall thickness and the mean ratio of inner bronchial to accompanying arterial diameter can be given as a function of lumen diameter. The mean wall thickness is defined as the difference between outer and inner bronchi radius.
it is not dependent on certain Hounsfield thresholds;
it works also on only partially closed bronchial walls;
it yields three-dimensional and subvoxel accuracies;
there is a clear criterion when to accept a measurement point: if the mean radial derivative curve shows a pronounced minimum following a pronounced maximum;
the measurement does not depend on the estimation of the local airway axis;
the same measurement principle can be applied for inner and outer airway wall as well as for artery diameters.
The search for an accompanying artery is conducted in a sphere of three times the radius of the outer bronchial wall around the center point 202. Within this search sphere the largest structure with vessel-morphology is identified. The measurements of the radii as well as the search for the accompanying artery can also be done in a two-dimensional disk perpendicular to the centerline of the airway.
The clinical parameters can be given for left and right lung separately or even per lobe. For example, a curve for each parameter for the principal longest bronchial centerlines can be given or a scatter plot with the clinical parameters as a function of distance to carina, which is a branching point in trachea, or as a function of bronchial lumen. Further, more different histograms of clinical parameters can be calculated. For example: the percentage of the length of the tracheobronchial tree which exhibits a certain lumen, wall thickness, the mean wall thickness and artery diameter ratio for bronchi pieces split up for different lumen ranges, etc.
The order in the described embodiments of the method of the current invention is not mandatory, a person skilled in the art may change the order of steps or perform steps concurrently using threading models, multi-processor systems or multiple processes without departing from the concept as intended by the current invention.
It should be noted that the above-mentioned embodiments illustrate rather than limit the invention, and that those skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word “comprising” does not exclude the presence of elements or steps other than those listed in a claim. The word “a” or “an” preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several distinct elements, and by means of a suitably programmed computer. In the system claims enumerating several means, several of these means can be embodied by one and the same item of computer readable software or hardware. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.
Number | Date | Country | Kind |
---|---|---|---|
04102883.8 | Jun 2004 | EP | regional |
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/IB05/51968 | 6/15/2005 | WO | 00 | 12/13/2006 |