Cerebrovascular disease (CVD), both clinical and subclinical, is a very significant health problem, especially in view of the increasing aging population. It is a major cause of dementia, individually or as an additive factor to other pathologies, such as Alzheimer's. CVD is also highly prevalent in diabetic populations, therefore measuring disease burden, progression, and response to treatments is very important for patient management. MRI is currently the best way to characterize the type and extent of brain lesions in CVD using imaging. However, characterization of CVD has largely relied on qualitative, subjective, and not easily reproducible methods of human expert-based interpretation, such as the Cardiovascular Health Study (CHS) grading of white matter lesions. Computer-based methods for measuring CVD offer great potential for measuring CVD, for many reasons: 1) they are quantitative;2) they are reproducible, and particularly suitable for longitudinal studies monitoring disease progression and response to candidate treatments;3) they are highly automated, thereby enabling the analysis of large amounts of data that are often acquired in large neuroimaging studies. The evaluation of disease progression has been particularly challenging, since typically relied on independent evaluations of baseline and follow- up scans then evaluating change, instead of maximizing our ability to evaluate change directly by jointly analyzing baseline and follow-up scans. This project aims to develop and validate quantitative image analysis methods that will enable the accurate and precise quantification of CVD and its progression over time. 4-dimensional image analysis methods will be developed, aiming to increase our sensitivity and accuracy in detecting subtle longitudinal changes of CVD and response to treatment. These methods will also utilize statistical atlases of normal brain structure, which will further improve the ability of automated methods to detect cerebrovascular lesions as being deviations from normal anatomy. Extensive validation will be performed using autopsy data, computational phantoms and expert definitions. Finally, application to one of the largest, to date, neuroimaging studies of diabetes and its clinical management, as well as to one of the most comprehensive longitudinal imaging studies of aging, will compare the sensitivity in detecting progression of CVD via this methodology, versus existing alternative image analysis approaches.
This project will develop image analysis methods for quantification of cerebrovascular disease, as well as its progression over time. 4-dimensional image segmentation methods will be used to obtain longitudinally consistent measurements of cerebrovascular pathology and its progression over time. Statistical atlases that capture normal variation of brain structure will further enhance the ability of automated computer methods to detect brain abnormalities as deviations from the normal range of variation. The methods will be tested on MRI scans from elderly individuals, as well as on MRI scans from diabetic patients with increased CVD.
|Erus, Guray; Zacharaki, Evangelia I; Davatzikos, Christos (2014) Individualized statistical learning from medical image databases: application to identification of brain lesions. Med Image Anal 18:542-54|
|Doshi, Jimit; Erus, Guray; Ou, Yangming et al. (2013) Multi-atlas skull-stripping. Acad Radiol 20:1566-76|