Abstract:

Deep neural networks have been successfully applied to medical image segmentation tasks, with their great potential to accelerate clinical workflows and facilitate large-scale studies. However, the performance of these deep segmentation models can be greatly impacted by changes in the data distribution due to scanner differences, varied imaging conditions as well as population shifts. To achieve satisfactory performance at deployment, these networks generally require massive labeled data collected from various domains (e.g., hospitals, scanners), which is rarely available in practice. In this talk, I will introduce our recent works on novel adversarial data augmentation algorithms to improve model generalization ability and robustness. The talk will cover two topics: a) input space adversarial data augmentation; b) feature-space adversarial data augmentation. 

Bio:

Chen (Cherise) Chen is currently a post-doc researcher in the Oxford BioMedIA group at the University of Oxford, as well as an honorary research fellow at Imperial College London. Chen received her Ph.D. from the Department of Computing at Imperial College London in January 2022 and then worked as a post-doc researcher in the same lab for one year. She also worked as a part-time research scientist at HeartFlow for six months in 2022. Her main research interests lie in the interdisciplinary area of AI and healthcare, with a focus on medical image analysis. Her doctoral thesis entitled “Improving the domain generalization and robustness of neural networks for medical imaging” was featured in ComputerVisionNews magazine. Chen has a good track record of winning challenges in medical image segmentation. She has won the Multi-sequence Cardiac MR Segmentation Challenge in 2019, and the Fetal Brain Tissue Annotation and Segmentation Challenge (FeTA) in 2022. She also served as the organizer of the workshop on Data Augmentation, Labeling, and Imperfections (DALI), MICCAI 2023, CMRxMotion challenge in the STACOM 2022 workshop, MICCAI 2022; Program committee in ECCV-MCV; Reviewer of IEEE TMI, JBHI, MedIA, MICCAI, IPMI, MIDL, etc. Chen has also been awarded as the IEEE TMI Gold-level distinguished reviewer (2020-2022). So far, she has published 28 papers in top conferences and journals on the topic of deep learning for medical data analysis, reaching 1200+ google scholar citations with an h-index of 14.