Home > Events > 2015 Seminars & Colloquia > CIPRIAN CRAINICEANU - Johns Hopkins University

CIPRIAN CRAINICEANU - Johns Hopkins University

Main Content

Statistical methods for stroke localization using clinical CT images
30 April 2015 from 4:00 PM to 5:00 PM
201 Thomas Bldg.
Add event to calendar

Qualitative radiologic description of intracranial hemorrhage (ICH) location has not been a robust clinical marker of ICH severity. Using advanced registration techniques on computed tomography (CT) brain scans, we provide detailed quantification of ICH location and its association with stroke outcomes. We analyzed 111 scans from 111 patients enrolled in the MISTIE (Minimally Invasive Surgery plus recombinant-tissue plasminogen activator for Intracerebral Evacuation) trial. We conclude that objective measures of ICH location and engagement using advanced CT imaging processing provide finer, more quantitative, and more predictive anatomic information than that provided by expert human readers. These findings suggest that important anatomic relationships may play biologically plausible roles in determining the severity of a particular ICH and can be estimated using our reproducible Statistical analytic platform. I will discuss methodological problems associated with this cross-sectional analysis of stroke CT brain imaging data and future plans for longitudinal analyses of the association between stroke location and disease severity. This will naturally lead to a discussion of the duality between the discovery and confirmatory phase of research. In particular, we propose voxel-wise analyses to generate brain areas that may be highly predictive of stroke outcomes, and a permutation test for the confirmatory phase. This will answer the question whether a procedure for selecting a highly predictive area of the brain produces statistically significant predictors, but comparing the relative importance of two or multiple selection procedures remains an open problem. 

Navigation for this Section