U-Net: A Versatile Deep Learning Architecture for Multi-Disease Detection

Imaging has revolutionized the medical domain due to its non-invasiveness and ability to capture internal organs accurately. The digitization of images led to the penetration of Artificial Intelligence, which improved the objectivity of analysis. The emergence of Convolution Neural Networks (CNN) allowed semantic segmentation of medical images, allowing pixel-level understanding. U-Net is a CNN, which is among the most robust architectures researchers use in state-of-the-art semantic segmentation methods. The chapter aims to discuss U-Net and its ability to work effectively across different imaging modalities. The chapter highlights variants of U-Nets and presents various data augmentation techniques to improve the performance. The chapter covers the segmentation of brain tumors, lung nodules, and liver lesions.