Predicting Tissue Outcome in Acute Ischemic Stroke Using Projection Pursuit Regression

Kartheeban Nagenthiraja¹, Kristjana Yr Jonsdottir¹, Leif stergaard¹, Kim Mouridsen¹

We have presented a flexible procedure for predicting final tissue outcome combining a dimension reduction method with efficient regression modeling of the resulting low-dimensional projections using smooth non-linear ridge functions. Increase in performance was observed for PPR notably in single-subject models, while performance in data pooled across patients was similar to LR, possibly owing to increased heterogeneity (reperfusion status, time of scan). We speculate that PPR in combination with techniques for identifying homogenous patient subgroups could further improve quantification of regional risk of infarct progression.