Progress in the field of deep learning has led to the development of neural network algorithms that can compete with, or even surpass, human performance in vision tasks, such as image classification or segmentation. This review provides a brief introduction to the building blocks of neural networks and explains why convolutional neural networks (CNNs) in particular excel at identifying relevant image features. In addition, it discusses how these CNNs learn to associate relevant image features with clinical features of interest, and how CNNs automatically segment structures of interest in an image volume. Finally, obstacles for wider application of these algorithms in scientific and clinical practice are addressed.
See Wiestler & Menze, Neuro-Oncology Advances