MICCAI 2022 Daily – Monday

over time so that we don’t need to select them. This new regularization method overcomes the label noise and the imbalanced learning problem . We can deal with both issues together. ” The work introduces two new evaluation benchmarks, training on noisy multi-label imbalanced chest X-ray training sets formed by Chest- Xray14 and CheXpert , and testing on clean multi-label datasets OpenI and PadChest . Compared with other methods, NVUM achieved state- of-the-art results. Previous works have studied the noisy labels and the imbalanced learning problem but have not tackled the two jointly. Fengbei says this is the first paper to propose training on both. Gustavo Carneiro’s team has six papers accepted at MICCAI this year – what does Fengbei think makes him particularly proud of this work? “ We’ve been working on this paper for two years, since my PhD started , and it’s not the first time we’ve submitted it to MICCAI, ” he reveals. “ We submitted a primary version last year but got rejected. However, we improved the paper quality and successfully made it this year. ” What was the trick? “ The main problem last year was that we didn’t consider the paper’s motivation. We only tried to solve the problem but didn’t state well enough why it’s so important for medical imaging that this problem is addressed. This year, we clearly highlighted the motivation in the paper’s first section . ” Fengbei says the team is already working on extensions and hopes doctors will warmly receive their contribution. “ We haven’t had any doctors see our work yet, but we’re hopeful they’ll be satisfied by our outcome, ” he tells us. “ Chest X-ray classification is an important task for doctors, who must train for a long time to be successful . If our model can help simplify the whole process, we’re happy to play a part in this. ” To learn more about Fengbei’s work, you are invited to visit Poster 1: Computer Aided Diagnosis I (In Person) today from 10:30-11:30. 21 DAILY MICCAI Monday Fengbei Liu