Figure 3

Analysis of misclassified patients. In the false positive case, the network seems to have picked up on remote infarcts that often resemble EPVS. In the false negative cases, there were less activations in the high range (> 7), i.e. activations were more homogenous.