摘要：Artificial intelligence (AI) has progressed rapidly in a variety of areas and achieved remarkable successes, ranging from autopiloting car to face recognition. Its applications in medicine have also attracted a lot of attentions. At the University of Texas Southwestern Medical Center, a group of researchers are dedicated to the development and application of deep learning (DL), the most common tool for AI nowadays, to solve problems in medical physics. This presentation will first give an overview of AI and DL. It will then present three example problems illustrating how we solve clinical medical physics problems with different complexities in cancer radiotherapy by employing DL and adapting it to the specific contexts to achieve different levels of machine intelligence. Examples will include 1) supervised learning using a deep neural network for applicator digitization in interstitial brachytherapy, 2) iterative DL for automatic sigmoid colon segmentation, and 3) deep-reinforcement learning for human-like automatic treatment planning. The last section of this presentation will discuss practical challenges when using DL to solve clinical problems. We will present the requirements to train the underlying deep neural networks, challenges to meet these requirements, and other mathematical and practical issues to consider.