Computer Vision News - March 2020

2 Summary AI in Medical Imaging 10 Klaus Maier-Hein leads the Division of Medical Image Computing at the German Cancer Research Center (DKFZ) and heads up the Section for Automated Image Analysis at Heidelberg University Hospital. He also served as area chair at MICCAI 2019 in Shenzhen. He speaks to us about his work and what it ’s like being the “go-to guy for medical image computing” in Heidelberg. Klaus and his team at DKFZ are working on the quantitative analysis of medical images, extracting information from image data for better diagnosis, prediction, and early detection of tumors. His work is mostly devoted to oncologic imaging – detecting and characterizing tumors and making predictions about tumor development and therapy. He has previously focused on brain tumors and prostate cancer. Klaus tel ls us that he would l ike to scale up the use of machine learning in medical appl ications. From his own experience and through speaking to col leagues, including those at MICCAI , he has identi f ied that data avai labi l ity is a common obstacle in the community. “How can we make data that is usual ly sitting in di fferent cl inics and hospitals more accessible? How can we enable federated set- ups and federated learning?” Klaus asks rhetorical ly. “We are actively bui lding the infrastructure that al lows us to interl ink di fferent cl inical centers in order to gain access to distributed data sources and are trying to leverage al l the data that is avai lable, even in distributed locations, to bui ld better models. That ’s a real ly thri l l ing development at the moment, which involves various methodological chal lenges, but also chal lenges on the infrastructural side.” He explains that people have been tackl ing the data avai labi l ity problem in di fferent ways. Some are working on methods that al low machine learning on fewer annotated datasets and combine, for example, unsupervised learning “How can we make data that is usually sitting in different clinics and hospitals more accessible?"