Multi-detector CT (MDCT) is rapidly becoming the primary imaging modality for detection and diagnosis of pulmonary embolism (PE). Its main limitations are its inability to reliably depict sub-segmental arteries and the large volume of image data that must be reviewed for each study. This application proposes to implement a stereographic display of appropriately segmented arterial trees, which can be manipulated in real time, with the intent of improving both the accuracy and efficiency of these studies. Most previous methods for segmenting pulmonary vessels are less than optimal in that, while they generally combine the eigenvectors of the Hessian matrix to derive a local scalar measure of cylindricity, in the process they lose information about the directions of the local curvatures. By exploiting all information in structure tensor fields derived from 3D datasets, and in particular by employing tensor voting methods to identify voxels comprising surfaces of vessels and bifurcations, we expect to significantly improve the segmentation process and the depiction of small arteries. A flexible mechanism will be developed for displaying various stereographic views of the segmented data in real time, at the option of the radiologist. Rendering methods, tailored specifically to viewing PE, will be developed. Multiple raycasting algorithms will be incorporated into the system because certain methods are better for detection while others are better for assessing a feature once it has been detected. Images comprised of local parameters used in segmenting vascular trees, or parameters that characterize statistical properties of vessel distributions, will be displayable at the readers'discretion. A retrospective LROC study (8 readers, 4 display modes, 100 cases) will be performed to evaluate the newly developed methods. This study will address performance and efficiency as well as certain psychophysical issues such as subjective acceptance of the display, speed of operation, pattern of gaze in 3D versus 2D, and the relative propensity of the various display modes to induce fatigue. To compensate for an imperfect gold standard for case verification, mixture distribution analysis will also be applied and compared to LROC results. This study should identify a set of images containing PEs that can be readily seen on stereo displays, but cannot be detected as easily on traditional displays, or vice versa - which should help clarify benefits of stereo for radiographic applications and provide useful information for making future refinements to the display.

Agency
National Institute of Health (NIH)
Institute
National Heart, Lung, and Blood Institute (NHLBI)
Type
Research Project (R01)
Project #
5R01HL084182-03
Application #
7568278
Study Section
Biomedical Imaging Technology Study Section (BMIT)
Program Officer
Moore, Timothy M
Project Start
2007-01-01
Project End
2010-12-31
Budget Start
2009-01-01
Budget End
2009-12-31
Support Year
3
Fiscal Year
2009
Total Cost
$371,250
Indirect Cost
Name
University of Pittsburgh
Department
Radiation-Diagnostic/Oncology
Type
Schools of Medicine
DUNS #
004514360
City
Pittsburgh
State
PA
Country
United States
Zip Code
15213
Pu, Jiantao; Fuhrman, Carl; Good, Walter F et al. (2011) A differential geometric approach to automated segmentation of human airway tree. IEEE Trans Med Imaging 30:266-78