Semantic segmentation was performed on 177 large histological images of re-endothelialized mouse lung vasculatures. Specifically, patch-based semantic segmentation algorithms were used to classify pixels corresponding to two classes: organ tissue, which includes lung and heart tissue; and ruptured and/or dilated vessels, which are abnormal vessels formed during re-endothelialization. Semantic segmentation is a potential means to automate the end-to-end analysis of these images, circumventing denoising and enhancement operations to visualize tissue and bypassing the manual diagnosis of ruptured and/or dilated vessels. To increase data quantity, images were compressed to sizes 1024 × 1024, 768 × 768, and 512 × 512 and then divided into nonoverlapping 256 × 256 patches. To benchmark the performance of the patch-based models, a vanilla model trained on complete images compressed to size 256 × 256 was also evaluated. The U-Net and LinkNet architectures were used to train and test each model using a data augmentation and transfer learning approach, and their results were ensembled. The loss of image context in the 3 × 3 and 4 × 4 patch-based models negatively impacted performance, generating many false positive predictions for target classes, whereas the low-image quantity of the vanilla model hindered performance. The 2 × 2 ensemble patch-based model returned the best performance, classifying organ tissue with a precision, recall, and intersection over union (IOU) of 88.0% ± 5.7%, 84.7% ± 9.2%, and 76.3% ± 10.9%, respectively, and classifying ruptured/dilated vessels with a precision, recall, and IOU of 78.4% ± 5.2%, 60.2% ± 11.4%, and 51.0% ± 8.4%, respectively.