Computer Vision News - April 2022

51 Deep Learning for Medical Imaging (DLMI) “ Most people say their goal isn’t to become an AI expert or develop new models; they want to apply known models to their data,” Pierre-Marc points out. “With AI in medical imaging, there are specific challenges you will not see in other fields. Medical data is huge, but databases can be small, with around 100 to 150 subjects . There are limits to AI, but evenwith small datasets, there is still plenty you can do if you do it right. That’s the overarching objective of the classes.” The week will feature a round-table discussion: ‘Why so much AI in research, why still so few AI in clinic?’ “ AI is everywhere in research, but if you walk into a hospital, you won’t see a lot of AI,” Pierre-Marc explains. “Why is that? How come AI doesn’t seem to percolate easily in a clinical setting? On the panel, we have a doctor, the CEO of a software development company, a medical expert, and a research scientist from industry, who will all give their views. “Here, it’s just one week, which is great for people in industry or students who only have a week off. It’s surprising how far we can get in just five days. Everything has been carefully chosen to give people a good starting point for navigating AI in medical imaging papers and playing around with libraries like PyTorch or TensorFlow .” The school has both theoretical classes and hands-on sessions. People who attend will receive an introduction to machine learning and the basics of deep learning before getting into the more advanced aspects of deep learning applied to medical imaging. Special presentations will be on popular topics, including generative and adversarial methods, explainability, and weakly supervised deep learning. With over 180 people already booked to attend on-site and virtually, the team has made arrangements with a GPU server farm to ensure participants have the resources needed to train state-of-the-art deep neural networks. Courtesy Emmanuel Roux - Lung Segmentation