Pulmonary embolism (PE) is one of leading cause of death in the United States if untreated. Prompt diagnosis and treatment can dramatically reduce the mortality rate and morbidity of the disease. Computed tomographic pulmonary angiography (CTPA) has been reported to be an effective means for clinical diagnosis of PE. Interpretation of a CT scan for PE demands extensive reading efforts from a radiologist who has to visually track a large number of vessels in the lungs to detect suspected PEs. Despite the efforts, the sensitivities were reported to range from 53% to 100%. Computer-aided diagnosis (CAD) can be a viable approach to improving the sensitivity and efficiency of PE detection in CTPA images, as well as reducing inter-observer variability. The overall goal of the proposed project is to develop a robust CAD system that can provide a systematic screening of PE and serve as a second opinion by automatically alerting the radiologists to suspicious locations on 2D slice and 3D volume rendering display of the CTPA images. We will develop advanced computer vision techniques to enhance the characteristics of vessels, automatically extract the pulmonary vessels, reconstruct the vessel tree, detect candidate PEs, differentiate PE from normal pulmonary structures, and identify the true PEs. The techniques will be specifically designed for analysis of the complex vascular structures on CTPA images.
The specific aims of this project include (1) collecting a large data set to develop and evaluate our CAD algorithms and systems, (2) establishing "gold standard" for performance evaluation, (3) developing robust pulmonary vessel segmentation methods, (4) developing robust pulmonary vessel tree reconstruction method to accurately track pulmonary vessels, trim veins and surrounding extensive lung diseases from vessel tree, and label reconstructed arterial tree, (5) developing and improving PE detection algorithms, including multi-prescreening method for the identification of suspicious PEs at different levels of artery branches, PE features extraction for development of classification methods, false positive reduction method based on feature analysis and fuzzy rule-based, linear, or neural network classifiers, (6) developing automatic PE index estimation method, (7) exploring performance evaluation methodology for computerized detection of PEs, and (8) performing observer ROC study to evaluate the effects of CAD on radiologists'accuracy in PE diagnosis.

Public Health Relevance

The relevance of this research to public health lies in the fact that there is substantial false-negative diagnosis of PEs. CAD will potentially reduce missed PEs and improve the chance of timely treatment of patients, thus reducing the mortality rate and speed up recovery from this condition.

Agency
National Institute of Health (NIH)
Institute
National Heart, Lung, and Blood Institute (NHLBI)
Type
Research Project (R01)
Project #
5R01HL092044-04
Application #
8315984
Study Section
Biomedical Imaging Technology Study Section (BMIT)
Program Officer
Moore, Timothy M
Project Start
2009-08-01
Project End
2014-07-31
Budget Start
2012-08-01
Budget End
2014-07-31
Support Year
4
Fiscal Year
2012
Total Cost
$470,300
Indirect Cost
$165,899
Name
University of Michigan Ann Arbor
Department
Radiation-Diagnostic/Oncology
Type
Schools of Medicine
DUNS #
073133571
City
Ann Arbor
State
MI
Country
United States
Zip Code
48109