Computer Vision News - June 2018

22 Challenge: Medical Decathlon Computer Vision News With Michela Antonelli Michela, can you tell us about the challenge at The aim of this challenge is to find the algorithm that can segment everything. With everything, I mean different types of organs/pathologies using different types of images. There are a lot of niche segmentation algorithms around, but what is missing is an algorithm that can be generalized and applied to different types of data set. This is the real challenge that we are proposing. You were personally charged with organizing the data, weren’t you? Yes, I collected all the data and it has been a very tough work! Clinicians sent me data in different formats with different names and I had to make it uniform and try to correct any mistakes. The whole data set has to be used by the same algorithm, so there has to be a common way of treating it to make running the algorithm as easy as possible. We managed to release the data set with the most open license possible, so people can use the data, change the data, and do whatever they want with it. It was a challenge to find hospitals that could give us data with this kind of license that could be released to the public in such an open way. Is there one specific hospital that was particularly cooperative you would like to name? Everyone has been really cooperative – I don’t want to forget someone! Amber Simpson at the Memorial Sloan Kettering Cancer Center. Bjoern Menze, Geert Litjens, and Patrick Christ , were all very helpful. Do you have any tips for the participants to the challenge? They should try to be as general as possible, because this challenge is divided into phases. In the first phase, they have seven tasks where they can train their algorithms and produce results. In the second phase, there are three mystery tasks that they don’t know anything about and they have Michela Antonelli is a research associate at University College London (UCL) in the Translational Imaging Group. She is also one of the organizers of this year’s Medical Segmentation Decathlon challenge, sponsored by NVIDIA, DeepMind and RSIP Vision. Challenge