Computer Vision News - March 2020

3 Summary Klaus Maier-Hein 1 only going to gain importance in the future. In medicine, we have such heterogeneous datasets and heterogeneous problems that we need to include al l these choices that we have to make going from one dataset to the other, from one data type to the other, from one organ to the other, and we have to systematical ly think about it. Only i f we do it in a systematicway and formal ize our choices, I think we can make real progress.” Looking to the future, Klaus hopes that methods wi l l learn faster and in a more eff icient and independent way. Less supervision. Less training data required. He hopes methods wi l l general ize better so they can be transferred from one dataset to another. For example, federated settings, where al l the di fferent qual ity types, scanner types, and aspects are covered so that real ly robust methods can be bui lt. He thinks it wi l l be important to focus on uncertainty handl ing because medicine is al l about handl ing and detecting uncertainties. The method has to know when it ’s not good, and when it has to inform the doctor whether it ’s certain or not about something. From a MICCAI perspective, when deal ing with images, he says another chal lenge to tackle in the future wi l l be how to integrate with other sources of information. For example, the imaging gives me a piece of information, but how does it combine with other pieces of information that are coming from