Massive amounts of efforts and energy were put into a type of residual Convolutional NN known as a U-net, which has had success in various 2d and 3d medical image segmentation tasks, finding irregularities or outliers, and performing general analysis. The architecture inputs an image, and uses a standard convolutional model, max pooling and downsampling to subsequent layers. the feature depth increases (usually doubles) as more depth is added. After multiple of these, the network then performs deconvolutional steps, taking the output of each prior downsample matching with the current feature map size and residually connecting the two. The network ends up looking like a "U" (hence the name). It is a very powerful model used in medical image analysis, prediction and segmentation. It also has seen use in the styles2paints model, proving its general purposeness.