Meeting Banner
Abstract #0757

Automated Stroke Disability Prediction & Mismatch Analysis by Employing Lesion Topography & Statistical Models

Roland Bammer1, Matus Straka1, Gregory W. Albers2, 3

1Center for Quantitative Neuroimaging, Department of Radiology, Stanford University, Stanford, CA, United States; 2Stanford Stroke Center, Department of Neurology, Stanford University, Stanford, CA, United States; 3on behalf of the DEFUSE investigators

DWI/PWI have been demonstrated to be reliable surrogate imaging markers for infarct core and at-risk tissue in acute stroke. Thus far, imaging-based prediction of clinical outcome has been primarily relied on overall lesion size and/or volumetric mismatch between stroke core and at-risk tissue. Here, we use a novel approach that employs importance-weighting to the mismatch analysis, where brain voxels contribute more or less to the scoring metric, based on their location and relative contribution to disability-based population statistics. Using a statistical model (i.e. stroke atlas), weights were derived from an acute stroke patient population and it could been shown that this topographic method predicts stroke disability extremely well.