Mark J. R. J. Bouts1, Ivo A. C. W. Tiebosch1, Rene Zwartbol1, Emily Hoogveld1, Ona Wu1,2, Rick M. Dijkhuizen1
1Image Sciences Institute, University Medical Center Utrecht, Utrecht, Netherlands; 2Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States
MRI-based predictive algorithms may perform favorably in identifying areas at risk of infarction following acute ischemic stroke, compared to the perfusion-diffusion mismatch. However, few studies have considered the ability of these methods to directly identify salvageable tissue. We evaluated four prediction methods in their potential to characterize salvageable tissue in a rat stroke model. A generalized linear model (glm) was tested against a support vector machine and two ensemble methods (adaboost, random forest). Our study shows that, under equal predictive performance of all methods, glm provides the greatest risk map contrast, enabling improved differentiation between irreversibly injured and potentially salvageable tissue after acute ischemic stroke.