Figure 2.

Verification of automatic vessel segmentation and arteriovenous classification. A, Representative input image, ground truth image, output image by automatic vessel segmentation, and merged image (top row). In merged images, yellow pixels are regarded as false negative results, pink pixels are regarded as false positive results, white pixels are regarded as true positive results, and black pixels are regarded as true negative results. Magnified images of the boxed area in each image above appear in the bottom row. B, Representative input image (left). Representative predicted arteriole image (middle). Red pixels represent the area belonging to the arteriole in ground truth and predicted as an arteriole by the deep learning program. Blue pixels represent the area belonging to the venule in ground truth, but predicted as an arteriole by the deep learning program. Representative predicted venule image (right). Blue pixels represent the area belonging to the venule in ground truth and predicted as a venule by the deep learning program. Red pixels represent the area belonging to the arteriole in ground truth, but predicted as a venule by the deep learning program.