Training neural network classifiers to delineate boundaries of tissue resection traditionally requires manually labeled MRIs. The process of manual labelling requires skill and time, and may be inconsistent among different people. Thus, researchers use a variety of techniques to augment their training dataset and get more training examples for their neural networks. One such augmentation method is to generate computer-simulated resections that visually look similar to the training set. We compare a classifier trained on these computer-simulated resections (RESSEG), a classifier trained on manual labels (DeepResection), and manual labels all against each other in a variety of metrics. The goal is to examine cases where either of these classifiers fail, and use this to improve upon the DeepResection classifier.