Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Dissteration.pdf (11.5 MB)

ETD Abstract Container

Abstract Header

STUDYING COMPUTATIONAL METHODS FOR BIOMEDICAL GEOMETRY EXTRACTION AND PATIENT SPECIFIC HEMODYNAMICS

wang, zhiqiang
http://rave.ohiolink.edu/etdc/view?acc_num=kent1493042299659479

Abstract Details

2017, PHD, Kent State University, College of Arts and Sciences / Department of Computer Science.
With the development of medical imaging and numerical simulation techniques, image based Patient Specific Computational Hemodynamics (PSCH) has become a powerful approach to non-invasively quantify vascular fluid mechanics in human arteries. However, most existing PSCH methods are currently impractical for real clinic applications due to time consuming computation. In clinic applications, an image based efficient PSCH approach is needed to offer Patient-Specific diagnosis information in a timely manner and also to make large population studies possible. Without these studies, the correlation between Patient-Specific clinical symptom and hemodynamic patterns cannot be assessed. In the image based PSCH processing procedure, there are three main parts responsible for the bulk of the computation. The first one is obtaining the anatomic geometry. The second is converting the geometry into a suitable mesh or grid for simulation. The third hotspot is the large numbers of calculations in hemodynamics simulation. To address above problems, in this thesis, some efficient algorithms have been proposed to speed up image based PSCH computations. First, a fully parallel numerical method is proposed to quickly extract anatomical geometry from medical images. This method can efficiently solve the level set equations of active contour based on Lattice Boltzmann model (LBM) for image segmentation. And a parallel distance field regularization algorithm is integrated to the LBM computing scheme to keep computation stable. This approach avoids external regularization which has been a major impediment to direct parallelization of level set evolution with LBM. It allows the whole computing process to be efficiently executed on Graphics Processing Unit (GPU). Further, this method can be incorporated with different image features for various image segmentation tasks. Second, a new method is proposed by utilizing fluid features, particularly the mean flow intensity, to extract the blood flow field in a target vessel from 4D flow MRI images which encode blood velocity information in vessels. This approach is computational efficient and robust to blood flow changes in a cardiac cycle even in relatively small arteries. The extracted velocity field can be used as inlet and outlet conditions facilitating the hemodynamics simulation and for evaluating simulation result. Moreover, the target artery wall deformation factors at a few spatial locations of the artery and time steps in one cardiac cycle can be estimated. These factors enable the generation of high quality and deforming artery walls by morphing the wall acquired from accurate but static Time of Flight (TOF) MRI images. The dynamic artery wall benefits blood flow visualization tasks and also has potential to be used as moving boundary for hemodynamic simulations. Third, after image segmentation, the anatomical geometry is implicitly represented by zero level set of distance field. Based on segmentation result, a method is first proposed to directly generate simulation grids for Volumetric Lattice Boltzmann Method (VLBM) based hemodynamic simulation. There is no surface reconstruction and mesh generation needed. Therefore, it can avoid extra computational cost and inaccuracy during the two transforms: from volume to mesh, then from mesh to volume grids. Further, surface normal direction at artery wall can also be directly estimated to calculate Wall Shear Stress (WSS) . Finally, VLBM has been GPU accelerated to simulate hemodynamics in human arteries by using a uniform computing scheme for both fluid and boundary grids. For traditional computational fluid simulation method, its boundary conditions have to be separately performed over boundary nodes from fluid nodes, where large number of branching operations are inefficient for GPU parallel computation due to its Single Instruction Multiple Data (SIMD) architecture. For complicated biomechanics structure, this situation would be worse, as there are a lot of boundary cells in the computation domain. The proposed parallel VLBM implementation does not need to distinguish fluid and boundary cells in the computation so that branching is minimized and the GPU kernel execution is accelerated.
Ye Zhao (Advisor)
Jun Li (Committee Member)
ChengChang Lu (Committee Co-Chair)
Robert Clements (Committee Chair)
Arden Rutten (Committee Member)
119 p.
Biomedical Engineering; Computer Science
Image segmentation, Hemodynamic simulation, Geometry processing, Parallel computing

Recommended Citations

Citations

  • wang, Z. (2017). STUDYING COMPUTATIONAL METHODS FOR BIOMEDICAL GEOMETRY EXTRACTION AND PATIENT SPECIFIC HEMODYNAMICS [Doctoral dissertation, Kent State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=kent1493042299659479

    APA Style (7th edition)

  • wang, zhiqiang. STUDYING COMPUTATIONAL METHODS FOR BIOMEDICAL GEOMETRY EXTRACTION AND PATIENT SPECIFIC HEMODYNAMICS. 2017. Kent State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=kent1493042299659479.

    MLA Style (8th edition)

  • wang, zhiqiang. "STUDYING COMPUTATIONAL METHODS FOR BIOMEDICAL GEOMETRY EXTRACTION AND PATIENT SPECIFIC HEMODYNAMICS." Doctoral dissertation, Kent State University, 2017. http://rave.ohiolink.edu/etdc/view?acc_num=kent1493042299659479

    Chicago Manual of Style (17th edition)

kent1493042299659479
257
© 2017, some rights reserved.Creative Commons License STUDYING COMPUTATIONAL METHODS FOR BIOMEDICAL GEOMETRY EXTRACTION AND PATIENT SPECIFIC HEMODYNAMICS by zhiqiang wang is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License. Based on a work at etd.ohiolink.edu.
This open access ETD is published by Kent State University and OhioLINK.