Personal tools
 

Detailed aims

Detailed project aims

  1. Establish efficient techniques for generating  subject-specific computational meshes for CFD analysis, including mesh construction of conducting airways from CT images and synthesized airways beyond the limitation of CT resolution using a volume-filling algorithm.
    1. Segmentation: Our airway segmentation method is based upon a combination of 3D region growing and 2D mathematical morphology (Aykac et al., 2003). The third generation of airway segmentation (Tschirren, 2003) uses fuzzy logic and has been demonstrated that airway trees can be successfully extracted from most clinical quality standard-dose CT data.
    2. Skeletonization and branch-point identification: The airway tree obtained by our segmentation method is skeletonized to identify the 3D centerlines of individual branches and branchpoint locations. A sequential 3D thinning algorithm (Palágyi et al., 2001, 2003) was customized for our application. After thinning, the skeleton is smoothed, pruned, and the complete tree converted into a graph structure using an adjacency list representation.
    3. Branch-point matching: Our airway-tree branch-point matching algorithm can find corresponding branch-points between two intra-subject scans in-vivo (Tschirren et al., 2002, 2003). The algorithm proved to be robust against false branches. Validation against human experts (FRC/TLC pairs, 10 normal and 7 diseased subjects) showed an average accuracy of 92.9% (average 22 matches per treepair). On average, 7.1% incorrect matches compares well with the expert inter-observer disagreement of 7.5%. The matching allows us to take images from multiple lung volumes, identify the segment correspondences, and feed this information to Tawhai’s model to build full dynamic airway trees.
    4. Anatomical labeling: We have developed an automatic method for anatomical labeling that is capable of assigning all 33 commonly used anatomical names to a human airway-tree (Tschirren et al., 2003). The method has been verified on 17 in-vivo TLC trees (10 normal and 7 diseased subjects).
    5. Branch morphometry and coloring: To identify individual airway tree branches, the segmented trees are transformed into a set of interconnected centerlines representing individual branches. Centerlines and associated airway structures are labeled, volume and surface area of each segment determined, and reproducibility assessed as shown in Fig. 10 (Palágyi et al., 2001, 2003a,b).
    6. Quantitative analysis of airway-tree segments: Fully automated quantitative analysis of airway-tree segments, using information from anatomical labeling, allows the linkage of measurements with the anatomical labels. The algorithm was verified on a physical phantom, with sub-voxel accuracy for all scan-directions and all airway sizes.
    7. Stereolithography: A volumetric rendering of the resultant segmented luminal space of the airway tree phantom is generated utilizing a marching cubes algorithm. We create triangular patches that divide the cube between regions within the isosurface and regions outside of airway tree. By connecting the patches from all cubes on the isosurface boundary, we get a triangulated surface representation of the airway tree. Each triangular patch consists of a single normal vector and three vertices. These triangular patchs are next converted to an STL file format.
    8. Surface remeshing for CFD mesh generation: If the triangulated surfaces created in the above step are not smooth for the CFD analysis owing to decreasing data resolution near the terminal branches, we will use the data fitting technique (Fernandez et al, 2004) developed by Dr. Tawhai’s group to smooth the surface geometry. This technique involves two steps: digitization and data fitting. Digitization is an automated process that uses CMGUI software (University of Auckland) to create discrete cloud points from CT/MR images. Data fitting uses an optimization process to fit a high-order cubic Hermite finite element mesh to the surface cloud data.
    9. The geometry acquired by CT imaging will be supplemented through modeling to represent the entire lung. The information derived above is critical for construction of entire lung geometry. The surface description of the lung lobes for each individual subject will be segmented from CT images and fitted to a high-order finite element volume mesh (Fernandez et al, 2004). For the same subject, a centerline model will be derived to fit their CT-segmented airways (Tawhai et al., 2000,2004a,b). The peripheral airways in the CT-based airway model are then used as starting locations for volume-filling.
  2. Integrate the custom developed 3D CFD model with the 1D gas transport model by developing an efficient algorithm to facilitate 3D to 1D coupling (large to small airways) or 1D to 3D coupling (bronchioles to alveolar ducts).
    1. Airway domain decomposition. Construct CFD models for different lung 'units'. i.e. trachea to fifth generation, fifth generation bronchus to generation 11, etc.
    2. Automatic surface mesh generation: An algorithm that can automatically generate 3D units 2-4 airway geometries and surface elements will be developed. These surface meshes can be imported into any mesh generator, such as GAMBIT®, for volume meshing. Our multiscale approach is uniquely designed to facilitate moving easily between different regions of interest in the airways. Because accurate flow predictions require large computational effort and resources, it is essential that meshes of the appropriate type can be constructed in an area of interest such that the 1D and 3D representations are spatially consistent. To create consistent 3D meshes of the non-segmented airways:
      1. A subject-specific 1D centerline model, including anatomically-consistent diameters, will be generated using the technique described by Tawhai et al. (2004). For the airway region of interest, high-order surface elements will be automatically generated to surround each centerline (left panel of Fig. 11). The surface elements will be exported to STL format, and the 3D CFD meshing will proceed as outlined in section C.2 and Specific Aim 1.
  3. Develop and experimentally validate a new predictive model of ventilation distribution by linking 3D CFD models to dynamic imaging of ventilation, via 1D flow models.
  4. Make available the coupling algorithms and databases to the research and clinical communities (two letters regarding intellectual property are attached as supplementary documents).