Figure 2.

Selected slice from the ADC parameter map of a patient treated using IV rtPA with overlay showing the ground-truth lesion segmentation (A) and the thresholded tissue outcome predictions of: the random decision forest classifier (B), the generalized linear model (C), and the k-nearest-neighbor classifier (D). Training sets used by the classifiers in generating these predictions incorporated voxel subsampling and normalization relative to the contralateral hemisphere.