Computer Vision News - March 2020

3 Summary Lung Nodule Database (LNDb( 3 US and China feature heavi ly, as expected, but Europe is wel l represented too. The team have publ ished an initial paper with dataset descriptions and some f indings regarding the annotation process. Can we get a sneak peek at the results? Gui lherme won’t give too much away: “One thing that is certain is that deep learning is completely dominating the f ield. Al l the approaches have deep learning associated with them. That ’s al l I can say.” But how do they ensure the data is rich and precise enough to enable deep learning activity? He explains: “Our data was assessed by at least four di fferent radiologists, but we do have nodules with di fferent agreement levels. We have nodules that only one radiologist marked, and we have nodules that al l radiologists marked. The idea is that participants train their systems knowing that there are various degrees of certainty. Some of the patients are more certain than others. One interesting thing that could affect system performance is that al l scans come from a local hospital , so they are adapted to the local real ity, and thus may not extrapolate for other populations.” Prizes for the best performing method in each part of the chal lenge wi l l be awarded at the 17th International Conference on Image Analysis and Recognition (ICIAR 2020) in Portugal in June. A total prize fund of $1,150 has been set aside. The grand chal lenge has been organized in partnership with the Institute for Systems and Computer Engineering, Technology and Science ( INESC TEC); Centro Hospitalar Universitário de São João; and the Faculty of Engineering and Faculty of Medicine at the University of Porto. After ICIAR 2020, al l partners wi l l be col laborating on a joint paper detai l ing the f inal results. "The idea is that participants train their systems knowing that there are various degrees of certainty"