While deep learning hitherto has achieved considerable success in medical image segmentation, existing methods suffer from significant performance
degradation under complex real-world clinical scenarios due to two main factors: (1) Label scarcity:
reliance on large-scale well-annotated datasets, and
(2) Domain shift: failure to apply well-trained
models trained on old labeled datasets to new unlabeled datasets with different data distributions. In
this paper, we introduce our recent works on labelefficient unsupervised domain adaptation to address both label scarcity and domain shift for crossdomain medical image analysis. We explore and
advance various label-efficient learning paradigms
with applications to medical image segmentation
for improving model generalization and efficiency.
This research is supported by core funding from: Institute for Infocomm Research, Science and Engineering Research Council
Grant Reference no. : N.A.